Financial compliance intelligent early warning system based on abnormal mode mining

Abstract

The invention discloses a financial compliance intelligent early warning system based on abnormal mode mining, and the system comprises a real-time data collection module which obtains transaction data from a financial system in real time and generates a multi-dimensional data structure; the data preprocessing module is used for performing data cleaning and standardization processing; the abnormal mode dynamic feature extraction module is used for generating an abnormal mode feature matrix based on the dynamic sliding time window; the feature optimization selection module is used for determining an optimal feature weight combination by adopting a chaotic ant lion optimization algorithm; the abnormal mode real-time identification module is used for analyzing the transaction data and identifying abnormal transactions; the intelligent dynamic early warning module automatically pushes warning information in real time; and the model continuous evolution module is used for updating algorithm parameters. According to the invention, intelligent and accurate financial abnormity early warning is realized, and the risk management and control capability is improved.

Description

Technical Field

[0001] This invention relates to the field of financial compliance risk monitoring technology, and in particular to a financial compliance intelligent early warning system based on anomaly pattern mining. Background Technology

[0002] Financial compliance risk monitoring technology plays a crucial role in the daily operations and management of enterprises. By promptly detecting and preventing abnormal or irregular transactions, it can effectively mitigate financial risks and improve management efficiency. In recent years, with the diversification and complexity of transaction scale and patterns, traditional financial compliance monitoring methods have gradually evolved from simple automated methods driven by manual review and fixed rules to intelligent analysis techniques based on data mining and pattern recognition. Especially in the fields of time series data analysis and abnormal pattern recognition, graph network structure analysis methods have gradually gained favor among those skilled in the art due to their unique advantages in describing transaction relationships and revealing the correlation of transaction behaviors.

[0003] Existing methods for detecting abnormal transactions often employ static threshold settings, rule engines, and anomaly identification based on traditional statistical methods. These solutions typically utilize pre-defined rule templates or static feature thresholds derived from historical data statistics, judging anomalies through simple comparisons or statistical deviations. However, such methods often fail to promptly detect new abnormal patterns or unknown abnormal transaction behaviors, especially when the manifestations of abnormal behavior exceed preset rules or statistical characteristics, easily leading to high false negative or false positive rates. Furthermore, because financial data exhibits significant temporal evolution, traditional methods lack dynamic response mechanisms to changes in data flow density, failing to effectively adjust detection thresholds and feature weights based on dynamic changes in data characteristics, thus impacting the real-time performance and accuracy of the monitoring system.

[0004] In recent years, to address the aforementioned issues, some technical solutions have proposed introducing dynamic temporal feature analysis and graph structure analysis methods. These solutions, by establishing temporal network graph structures and combining them with machine learning models to mine anomaly patterns, have achieved some progress. However, these solutions generally employ random or simple heuristic algorithms in feature selection and threshold optimization, making it difficult to accurately locate the optimal anomaly detection parameters. This limits the system's detection efficiency and accuracy, resulting in insufficient system stability. Furthermore, most current anomaly pattern recognition results lack structured descriptions and risk level assessment mechanisms, failing to intuitively present the overall picture and risk path of abnormal transactions, thus posing challenges to compliance managers' risk decision-making.

[0005] Therefore, how to provide a financial compliance intelligent early warning system based on anomaly pattern mining is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a financial compliance intelligent early warning system based on anomaly pattern mining. This invention has the technical advantages of real-time and accurate detection, low false alarm and false negative rates, and intuitive risk assessment.

[0007] A financial compliance intelligent early warning system based on anomaly pattern mining according to an embodiment of the present invention includes the following modules:

[0008] The real-time data acquisition module is used to acquire transaction data from the financial system in real time and generate a multi-dimensional data structure containing transaction amount, transaction time, account information, and related party information.

[0009] The data preprocessing module is used to clean, impute missing values, and standardize multidimensional data structures to form a financial compliance monitoring data set.

[0010] The abnormal pattern dynamic feature extraction module is used to transform the financial compliance monitoring data set into a network structure diagram composed of transaction sequences based on a dynamic sliding time window, and extract the temporal features, frequency features and network correlation features of transaction events to generate an abnormal pattern feature matrix.

[0011] The feature optimization and selection module is used to input the abnormal pattern feature matrix into the chaotic antlion optimization algorithm, and then obtain the best feature weight combination and abnormal threshold for abnormal pattern recognition through the chaotic dynamic search mechanism of hunting and random walk.

[0012] The real-time anomaly pattern identification module is used to analyze and monitor financial transaction data in real time by utilizing the optimal feature weight combination and anomaly threshold, and generate anomaly pattern identification results when it detects transaction behavior that deviates from the normal pattern.

[0013] The intelligent dynamic early warning module is used to automatically trigger and push warning information to compliance management personnel in real time after receiving the abnormal pattern recognition results;

[0014] The model continuous evolution module is used to receive and store warning false alarms and missed alarms after manual confirmation, and update the search space and parameter configuration of the chaotic antlion optimization algorithm.

[0015] Optionally, modules can be integrated using the following methods:

[0016] S1. Obtain transaction data in real time from the financial system interface and generate a multi-dimensional data structure in a unified format;

[0017] S2. Through data cleaning rules and data standardization algorithms, multi-dimensional data structures are processed in a unified manner to form a set of financial compliance monitoring data.

[0018] S3. Based on the preset dynamic sliding time window rules, the financial compliance monitoring data set is sequentially transformed into a serialized network structure diagram, and the transaction event time sequence features, frequency features and network correlation features are extracted from the network structure diagram node by node, and the abnormal pattern feature matrix is ​​output.

[0019] S4. Call the chaotic antlion optimization algorithm built into the feature optimization selection module. Through the chaotic dynamic search mechanism that combines hunting position selection and random walk trajectory optimization, automatically calculate and determine the best feature weight combination and abnormal threshold in the abnormal pattern feature matrix for abnormal pattern recognition.

[0020] S5. Utilize the optimal feature weight combination and anomaly threshold to continuously analyze and monitor financial transaction data in the financial compliance monitoring dataset. When transaction data is found to deviate significantly from the normal pattern, output the structured anomaly pattern identification result.

[0021] S6. Based on predefined warning rules, warning messages are automatically generated in real time according to the structured anomaly pattern recognition results, and the warning push interface is called to send them to the terminal devices of designated compliance management personnel in an instant.

[0022] S7. Based on the warning false alarms and omissions confirmed by compliance management personnel, the data is automatically stored in the data buffer, and the search space and parameter configuration of the chaotic antlion optimization algorithm are dynamically updated according to the predefined continuous evolution rules.

[0023] Optionally, S1 specifically includes:

[0024] S11. Construct a dual-channel asynchronous data acquisition channel between the financial system interface and the intelligent early warning system. Channel 1 adopts a periodic active polling mode, while Channel 2 adopts a real-time triggering mode based on transaction events. The two channels run in parallel and independently, and each is set with an independent acquisition clock.

[0025] S12. Embed a unique transaction sequence tag and a channel source tag into each transaction data acquired asynchronously through dual channels, and establish and store a one-to-one mapping relationship between the tag and the transaction data.

[0026] S13. Real-time call to the asynchronous data cross-validation mechanism, through the mapping relationship of the identification tags, to compare and verify the data of the same transaction event obtained by Channel 1 and Channel 2 respectively, automatically identify and mark the existence of field data differences, data missing and duplicate reception;

[0027] S14. Based on the cross-validation mechanism, automatically generate a data consistency verification report, and use data compensation and deduplication rules to compensate for the difference of missing data, intelligently merge duplicate data, and dynamically and adaptively repair transaction data.

[0028] S15. Based on the automatically repaired transaction data, construct a three-dimensional transaction attribute label set including account type classification labels, transaction time window labels, and real-time dynamic direction labels of fund flow, and perform labeling processing on the transaction data.

[0029] S16. After the transaction data has been processed by three-dimensional transaction attribute tagging, it is encapsulated one by one according to the unified data structure standard rules to generate a multi-dimensional data structure.

[0030] Optionally, S2 specifically includes:

[0031] S21. A dynamic field constraint mechanism is adopted to detect the changing trend of transaction amount values ​​and account classification labels in multidimensional data structures in real time, and to automatically update the outlier identification threshold to identify data with abnormal field values.

[0032] S22. Construct a multi-dimensional data structure association graph based on the time window label of the trading period and the real-time dynamic direction label of the fund flow, and identify data units with missing or incomplete association relationships;

[0033] S23. By comparing the topological position of the multidimensional data structure association graph with the identified abnormal data units in context, extract data reference values ​​from the upstream and downstream data units adjacent to the topological position, and dynamically correct the abnormal data units.

[0034] S24. Based on the historical distribution of account type classification labels and transaction amount data, generate a personalized data integrity model, detect and automatically fill in data units with missing or incomplete relationships;

[0035] S25. By constructing a dynamic feature space for transaction data, adaptive nonlinear mapping is performed on the corrected and filled transaction data to generate standardized data with uniformly distributed data features.

[0036] S26. Utilize the time-series data balanced distribution rules to verify the time-series continuity and integrity of standardized data, and generate a financial compliance monitoring data set.

[0037] Optionally, S3 specifically includes:

[0038] S31. Based on the statistical indicators of data density and transaction frequency of each data unit in the financial compliance monitoring data set, calculate and adjust the window width and sliding step of the dynamic sliding time window to form a non-fixed length sliding time window that dynamically adapts to the data flow density.

[0039] S32. For each set of financial compliance monitoring data within a non-fixed length sliding time window, analyze each transaction data item by item in the order of the occurrence of the transaction events, determine the relationship between the transaction entity node and the transaction, and establish a dynamic time-series network structure diagram of the transaction events.

[0040] S33. Based on the positional centrality and historical transaction activity of each node in the dynamic time-series network structure graph, dynamically calculate the node transaction frequency weighting coefficient, and use the node transaction frequency weighting coefficient to adjust the original transaction frequency of the node node by node to generate the node activity weighted transaction frequency feature.

[0041] S34. For each node, record and extract the occurrence time of node transaction events in the dynamic temporal network structure diagram, calculate the time interval difference between adjacent transaction events, and form a node transaction temporal feature sequence that is sensitive to transaction intervals.

[0042] S35. Based on the dual constraints of the order of transaction occurrence and the difference in transaction amount between nodes, evaluate and eliminate invalid transaction associations, retain only valid node connections with temporal causal relationship and amount matching relationship, calculate the number of connections between a node and all valid associated nodes, and generate node valid transaction network association characteristics.

[0043] S36. Based on the dynamic changes in the transaction frequency characteristics, transaction timing characteristics, and network association characteristics of nodes within the current window data, calculate the dynamic fusion weight coefficients of various features. Through the adaptive combination rules of the dynamic fusion weight coefficients, fuse all features of the nodes to construct an abnormal pattern feature matrix with dynamic feature weight adjustment.

[0044] Optionally, S4 specifically includes:

[0045] S41. Set the initial search space for the chaotic antlion optimization algorithm, and use the transaction frequency feature, transaction time sequence feature and network correlation feature in the abnormal mode feature matrix as the feature dimensions to be optimized.

[0046] S42. According to the logistic chaotic mapping equation, initialize the initial position of the ant in the algorithm;

[0047] S43. In each iteration, a local search domain is constructed based on the current hunting position of the antlion. The ant's walking trajectory in the local search domain is randomly generated based on the chaotic mapping. The walking trajectory is defined as a candidate solution of the feature weight combination.

[0048] S44. The classification error rate of the abnormal mode feature matrix is ​​used as the fitness function to evaluate the fitness of candidate solutions in real time.

[0049] S45. Record the fitness function evaluation results, select the feature weight combination with the highest fitness and the corresponding anomaly threshold as the optimal candidate solution for the current iteration, and dynamically adjust the antlion hunting position and local search domain according to the evaluation results.

[0050] S46. Set the algorithm iteration termination condition as the change in the fitness function over M consecutive iterations being less than the convergence threshold ∈. After reaching the termination condition, output the final determined optimal feature weight combination and anomaly threshold.

[0051] Optionally, S43 specifically includes:

[0052] S431. Based on the current hunting position of the antlion and the changing trend of the hunting position in the historical iteration, calculate and dynamically adjust the center position of the local search domain, and determine the center of the local search domain by the weighted historical average value of the position change vector.

[0053] S432. Based on the concentration of the global optimal position on the feature dimension, adjust the range coefficient of the local search domain in real time, and dynamically expand and shrink the local search domain.

[0054] S433. Based on the mapping sequence of the logistic chaotic mapping equation, perform multi-scale reconstruction, perform multi-scale hierarchical sampling on the mapping sequence, generate chaotic initial position sequences at different scales, and perform multi-scale coverage on the local search domain.

[0055] S434. Merge the multi-scale chaotic initial position sequence with the upper and lower boundaries of the current local search domain to generate an initial wandering point set with multi-scale characteristics, and construct the wandering trajectory step by step according to the priority of the mapping sequence scale.

[0056] S435. Utilize the historical difference between the antlion and the ant's position to calculate the dynamic trajectory update direction in real time, form a direction guiding factor, and combine it with the chaotic mapping sequence to generate an ant walking trajectory with direction preference.

[0057] S436. Record and extract the high-fitness segments of the walking trajectory in each iteration, and merge them in real time according to the degree of segment overlap to form candidate solutions of feature weight combination.

[0058] Optionally, S5 specifically includes:

[0059] S51. Based on the optimal feature weight combination, perform feature weight weighting processing on each dimension of each transaction data in the financial compliance monitoring data set, and calculate and obtain the comprehensive abnormal feature score of each transaction data.

[0060] S52. Using the anomaly threshold, compare the comprehensive anomaly feature score of each transaction data with the numerical value of the anomaly threshold to determine and mark the anomaly status.

[0061] S53. For transaction data marked as abnormal, extract the account type classification label, transaction time window label, real-time dynamic direction label of fund flow, transaction amount value and corresponding comprehensive abnormal feature score in chronological order to form a structured abnormal transaction information record.

[0062] S54. Based on the account type classification tags and real-time dynamic direction tags of fund flow in the structured abnormal transaction information records, perform correlation path analysis on the data flow of abnormal transactions to determine the associated account paths of abnormal transactions.

[0063] S55. For each abnormal transaction's associated account path, analyze each path and determine the transaction risk level of each path based on the number of account nodes in the path, the transaction amount, and the comprehensive abnormal characteristic score.

[0064] S56. Integrate the structured information records of abnormal transactions, the associated account paths, and the determined transaction risk levels to form the final structured abnormal pattern recognition result.

[0065] Optionally, S6 specifically includes:

[0066] S61. Calculate the dynamically adjusted warning trigger priority based on the transaction risk level and the historical anomaly handling records of the transaction account in the structured anomaly pattern recognition results.

[0067] S62. By combining dynamically adjusted warning trigger priorities with predefined warning message generation templates, personalized warning message content is dynamically generated based on the differences in the characteristics of abnormal transaction account type classification tags, transaction time window tags, and real-time dynamic direction tags of fund flow.

[0068] S63. Based on the real-time status and historical response time records of the terminal devices of the designated compliance management personnel, select and adjust the priority level and push mechanism of the message push interface in real time.

[0069] S64. Utilize the priority matching results between warning messages and push interfaces to automatically and in real time generate enhanced standardized data messages containing abnormal transaction data feature information and suggested response strategies.

[0070] S65. Monitor the network communication quality of the terminal devices of designated compliance management personnel in real time during message push, dynamically adjust the segmentation strategy and transmission batch of push data packets, and optimize the success rate of warning message push.

[0071] S66. Based on the real-time response status and reading confirmation time of the terminal device, automatically update the push priority, push strategy and the response model of the terminal device for warning messages, and optimize the push process for the next warning message in real time.

[0072] Optionally, S7 specifically includes:

[0073] S71. Acquire and record warning false alarms and omissions after manual confirmation by compliance management personnel, and store them in the data buffer according to the comprehensive abnormal feature score of transaction data, account type classification label, transaction time window label, and real-time dynamic direction label of fund flow.

[0074] S72. Based on the false alarms and missed alarms in the data buffer, analyze the deviation of each transaction data in the dimension of abnormal characteristics, and determine the error distribution in transaction frequency characteristics, transaction time sequence characteristics and network correlation characteristics.

[0075] S73. Based on the determined error distribution, calculate and generate the adjustment factor corresponding to the feature dimension. The adjustment factor is used to update the initial search space boundary of the chaotic antlion optimization algorithm.

[0076] S74. Based on the historical statistics and frequency of false alarms and omissions in transaction data, calculate the dynamic adjustment amount used to update the chaotic mapping control parameters in the chaotic antlion optimization algorithm.

[0077] S75. Update the upper and lower boundary values ​​of the initial search space of the chaotic antlion optimization algorithm in real time based on the adjustment factor, and synchronously input the dynamically adjusted chaotic mapping control parameters into the chaotic antlion optimization algorithm.

[0078] S76. Based on the predefined continuous evolution rules, automatically update the initial search space boundary values ​​and dynamically adjust the chaotic mapping control parameters.

[0079] The beneficial effects of this invention are:

[0080] (1) This invention effectively reduces the false alarm rate and false negative rate in the abnormal transaction identification process by adaptively adjusting the window width and step size based on a dynamic sliding time window, and accurately identifying abnormal patterns by using feature weight combination and abnormal threshold. This significantly enhances the real-time performance and accuracy of abnormal monitoring.

[0081] (2) This invention uses the Chaotic Antlion Optimization Algorithm to determine the optimal feature weight combination and anomaly threshold in real time, thereby achieving efficient and automatic optimization of anomaly pattern feature selection and anomaly threshold, significantly improving the efficiency and accuracy of anomaly identification, and demonstrating better adaptability in complex and dynamic financial transaction environments.

[0082] (3) In terms of risk level assessment and early warning of abnormal transactions, this invention effectively solves the shortcomings of existing technologies in lacking risk path assessment and structured output by tracing transaction paths and analyzing risk levels. It breaks through the technical bottleneck of traditional abnormal detection methods in locating risks of related accounts, realizes refined and intuitive monitoring of financial abnormalities, and thus effectively improves the risk decision-making efficiency of financial compliance management. Attached Figure Description

[0083] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0084] Figure 1 This is a system architecture diagram of a financial compliance intelligent early warning system based on anomaly pattern mining proposed in this invention.

[0085] Figure 2 This is a dynamic sliding time window feature extraction diagram for a financial compliance intelligent early warning system based on anomaly pattern mining proposed in this invention.

[0086] Figure 3 This is a flowchart of the chaotic antlion optimization algorithm for a financial compliance intelligent early warning system based on anomaly pattern mining proposed in this invention. Detailed Implementation

[0087] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0088] refer to Figures 1-3 A financial compliance intelligent early warning system based on anomaly pattern mining includes the following modules:

[0089] The real-time data acquisition module is used to acquire transaction data from the financial system in real time and generate a multi-dimensional data structure containing transaction amount, transaction time, account information, and related party information.

[0090] The data preprocessing module is used to clean, impute missing values, and standardize multidimensional data structures to form a financial compliance monitoring data set.

[0091] The abnormal pattern dynamic feature extraction module is used to transform the financial compliance monitoring data set into a network structure diagram composed of transaction sequences based on a dynamic sliding time window, and extract the temporal features, frequency features and network correlation features of transaction events to generate an abnormal pattern feature matrix.

[0092] The feature optimization and selection module is used to input the abnormal pattern feature matrix into the chaotic antlion optimization algorithm, and then obtain the best feature weight combination and abnormal threshold for abnormal pattern recognition through the chaotic dynamic search mechanism of hunting and random walk.

[0093] The real-time anomaly pattern identification module is used to analyze and monitor financial transaction data in real time by utilizing the optimal feature weight combination and anomaly threshold, and generate anomaly pattern identification results when it detects transaction behavior that deviates from the normal pattern.

[0094] The intelligent dynamic early warning module is used to automatically trigger and push warning information to compliance management personnel in real time after receiving the abnormal pattern recognition results;

[0095] The model continuous evolution module is used to receive and store warning false alarms and missed alarms after manual confirmation, and update the search space and parameter configuration of the chaotic antlion optimization algorithm.

[0096] By collecting financial transaction data in real time and constructing a multi-dimensional data structure, extracting anomaly pattern feature matrices based on a dynamic sliding time window, and using the Chaotic Antlion Optimization Algorithm to determine the optimal feature weight combination and anomaly threshold, abnormal transactions can be analyzed and identified in real time and automatically triggered to trigger early warnings. At the same time, the algorithm parameters are dynamically updated and optimized based on feedback data from compliance personnel, thereby improving the accuracy and real-time performance of financial anomaly identification, effectively reducing false alarm and false negative rates, and enhancing the monitoring and control capabilities of financial risks.

[0097] In this embodiment, the modules are interconnected using the following method:

[0098] S1. Obtain transaction data in real time from the financial system interface and generate a multi-dimensional data structure in a unified format;

[0099] S2. Through data cleaning rules and data standardization algorithms, multi-dimensional data structures are processed in a unified manner to form a set of financial compliance monitoring data.

[0100] S3. Based on the preset dynamic sliding time window rules, the financial compliance monitoring data set is sequentially transformed into a serialized network structure diagram, and the transaction event time sequence features, frequency features and network correlation features are extracted from the network structure diagram node by node, and the abnormal pattern feature matrix is ​​output.

[0101] S4. Call the chaotic antlion optimization algorithm built into the feature optimization selection module. Through the chaotic dynamic search mechanism that combines hunting position selection and random walk trajectory optimization, automatically calculate and determine the best feature weight combination and abnormal threshold in the abnormal pattern feature matrix for abnormal pattern recognition.

[0102] S5. Utilize the optimal feature weight combination and anomaly threshold to continuously analyze and monitor financial transaction data in the financial compliance monitoring dataset. When transaction data is found to deviate significantly from the normal pattern, output the structured anomaly pattern identification result.

[0103] S6. Based on predefined warning rules, warning messages are automatically generated in real time according to the structured anomaly pattern recognition results, and the warning push interface is called to send them to the terminal devices of designated compliance management personnel in an instant.

[0104] S7. Based on the warning false alarms and omissions confirmed by compliance management personnel, the data is automatically stored in the data buffer, and the search space and parameter configuration of the chaotic antlion optimization algorithm are dynamically updated according to the predefined continuous evolution rules.

[0105] By acquiring financial transaction data in real time and processing it in a unified format, and using a dynamic sliding time window to extract the timing, frequency, and network correlation features of transaction events, the Chaotic Antlion Optimization Algorithm is used to automatically determine the feature weight combination and anomaly threshold for anomaly pattern recognition. The algorithm parameters are continuously optimized by combining the feedback from compliance management personnel on false alarms and missed alarms. This significantly improves the accuracy and real-time performance of anomaly transaction detection, reduces false alarms and missed alarms, and enhances the overall monitoring and management capabilities of financial compliance risks.

[0106] In this embodiment, S1 specifically includes:

[0107] S11. Construct a dual-channel asynchronous data acquisition channel between the financial system interface and the intelligent early warning system. Channel 1 adopts a periodic active polling mode, while Channel 2 adopts a real-time triggering mode based on transaction events. The two channels run in parallel and independently, and each is set with an independent acquisition clock.

[0108] S12. Embed a unique transaction sequence tag and a channel source tag into each transaction data acquired asynchronously through dual channels, and establish and store a one-to-one mapping relationship between the tag and the transaction data.

[0109] S13. Real-time call to the asynchronous data cross-validation mechanism, through the mapping relationship of the identification tags, to compare and verify the data of the same transaction event obtained by Channel 1 and Channel 2 respectively, automatically identify and mark the existence of field data differences, data missing and duplicate reception;

[0110] S14. Based on the cross-validation mechanism, automatically generate a data consistency verification report, and use data compensation and deduplication rules to compensate for the difference of missing data, intelligently merge duplicate data, and dynamically and adaptively repair transaction data.

[0111] S15. Based on the automatically repaired transaction data, construct a three-dimensional transaction attribute label set including account type classification labels, transaction time window labels, and real-time dynamic direction labels of fund flow, and perform labeling processing on the transaction data.

[0112] S16. After the transaction data has been processed by three-dimensional transaction attribute tagging, it is encapsulated one by one according to the unified data structure standard rules to generate a multi-dimensional data structure.

[0113] By constructing a dual-channel asynchronous data acquisition system, real-time cross-validation of data consistency between channels is performed. Data discrepancies, missing data, and duplicate data are automatically identified. Through discrepancy compensation and intelligent merging, data is dynamically and adaptively repaired to form a standardized multidimensional data structure with consistent structure. This effectively improves the accuracy and completeness of financial transaction data collection, providing high-quality data support for subsequent accurate anomaly pattern identification.

[0114] In this embodiment, S2 specifically includes:

[0115] S21. A dynamic field constraint mechanism is adopted to detect the changing trend of transaction amount values ​​and account classification labels in multidimensional data structures in real time, and to automatically update the outlier identification threshold to identify data with abnormal field values.

[0116] S22. Construct a multi-dimensional data structure association graph based on the time window label of the trading period and the real-time dynamic direction label of the fund flow, and identify data units with missing or incomplete association relationships;

[0117] S23. By comparing the topological position of the multidimensional data structure association graph with the identified abnormal data units in context, extract data reference values ​​from the upstream and downstream data units adjacent to the topological position, and dynamically correct the abnormal data units.

[0118] S24. Based on the historical distribution of account type classification labels and transaction amount data, generate a personalized data integrity model, detect and automatically fill in data units with missing or incomplete relationships;

[0119] S25. By constructing a dynamic feature space for transaction data, adaptive nonlinear mapping is performed on the corrected and filled transaction data to generate standardized data with uniformly distributed data features.

[0120] S26. Utilize the time-series data balanced distribution rules to verify the time-series continuity and integrity of standardized data, and generate a financial compliance monitoring data set.

[0121] By using a dynamic field constraint mechanism to monitor data anomalies in real time and automatically update the anomaly identification threshold, and by using a multi-dimensional data structure association graph to identify and dynamically repair missing and incomplete data, and by combining personalized data integrity patterns and dynamic feature spaces to generate standardized data, the temporal continuity and integrity of the data are ensured. This significantly improves the quality and consistency of the financial compliance monitoring data set and effectively guarantees the accuracy of abnormal transaction identification.

[0122] In this embodiment, S3 specifically includes:

[0123] S31. Based on the statistical indicators of data density and transaction frequency of each data unit in the financial compliance monitoring data set, calculate and adjust the window width and sliding step of the dynamic sliding time window to form a non-fixed length sliding time window that dynamically adapts to the data flow density.

[0124] S32. For each set of financial compliance monitoring data within a non-fixed length sliding time window, analyze each transaction data item by item in the order of the occurrence of the transaction events, determine the relationship between the transaction entity node and the transaction, and establish a dynamic time-series network structure diagram of the transaction events.

[0125] S33. Based on the positional centrality and historical transaction activity of each node in the dynamic time-series network structure graph, dynamically calculate the node transaction frequency weighting coefficient, and use the node transaction frequency weighting coefficient to adjust the original transaction frequency of the node node by node to generate the node activity weighted transaction frequency feature.

[0126] S34. For each node, record and extract the occurrence time of node transaction events in the dynamic temporal network structure diagram, calculate the time interval difference between adjacent transaction events, and form a node transaction temporal feature sequence that is sensitive to transaction intervals.

[0127] S35. Based on the dual constraints of the order of transaction occurrence and the difference in transaction amount between nodes, evaluate and eliminate invalid transaction associations, retain only valid node connections with temporal causal relationship and amount matching relationship, calculate the number of connections between a node and all valid associated nodes, and generate node valid transaction network association characteristics.

[0128] S36. Based on the dynamic changes in the transaction frequency characteristics, transaction timing characteristics, and network association characteristics of nodes within the current window data, calculate the dynamic fusion weight coefficients of various features. Through the adaptive combination rules of the dynamic fusion weight coefficients, fuse all features of the nodes to construct an abnormal pattern feature matrix with dynamic feature weight adjustment.

[0129] By adaptively adjusting the dynamic sliding time window to adapt to changes in data flow density, a dynamic temporal network structure diagram of transaction events is constructed. Node activity weighted features are generated by dynamically weighting node position centrality and transaction frequency. The transaction temporal feature sequence and effective network association features are accurately calculated. Furthermore, various feature weights are dynamically fused and adjusted in real time to construct an abnormal pattern feature matrix. This effectively improves the sensitivity and accuracy of abnormal transaction identification and enhances the system's adaptability to complex financial transaction environments.

[0130] In this embodiment, S4 specifically includes:

[0131] S41. Set the initial search space for the chaotic antlion optimization algorithm, and use the transaction frequency feature, transaction time sequence feature and network correlation feature in the abnormal mode feature matrix as the feature dimensions to be optimized.

[0132] S42. According to the logistic chaotic mapping equation, initialize the initial position of the ant in the algorithm;

[0133] S43. In each iteration, a local search domain is constructed based on the current hunting position of the antlion. The ant's walking trajectory in the local search domain is randomly generated based on the chaotic mapping. The walking trajectory is defined as a candidate solution of the feature weight combination.

[0134] S44. The classification error rate of the abnormal mode feature matrix is ​​used as the fitness function to evaluate the fitness of candidate solutions in real time.

[0135] S45. Record the fitness function evaluation results, select the feature weight combination with the highest fitness and the corresponding anomaly threshold as the optimal candidate solution for the current iteration, and dynamically adjust the antlion hunting position and local search domain according to the evaluation results.

[0136] S46. Set the algorithm iteration termination condition as the change in the fitness function over M consecutive iterations being less than the convergence threshold ∈. After reaching the termination condition, output the final determined optimal feature weight combination and anomaly threshold.

[0137] By setting the initial search space of the chaotic antlion optimization algorithm and determining the initial position of the ants by combining the chaotic mapping sequence, the algorithm iteratively generates local optimal candidate solutions using the antlion hunting and random chaotic dynamic search mechanism. Furthermore, it uses the classification error rate of the abnormal pattern feature matrix to evaluate the fitness in real time, and dynamically adjusts and optimizes the combination of feature weights and the abnormal threshold. This effectively improves the parameter optimization accuracy and efficiency of abnormal pattern recognition, and enhances the system's anomaly detection accuracy and real-time response capability.

[0138] In this embodiment, S43 specifically includes:

[0139] S431. Based on the current hunting position of the antlion and the changing trend of the hunting position in the historical iteration, calculate and dynamically adjust the center position of the local search domain, and determine the center of the local search domain by the weighted historical average value of the position change vector.

[0140] S432. Based on the concentration of the global optimal position on the feature dimension, adjust the range coefficient of the local search domain in real time, and dynamically expand and shrink the local search domain.

[0141] S433. Based on the mapping sequence of the logistic chaotic mapping equation, perform multi-scale reconstruction, perform multi-scale hierarchical sampling on the mapping sequence, generate chaotic initial position sequences at different scales, and perform multi-scale coverage on the local search domain.

[0142] S434. Merge the multi-scale chaotic initial position sequence with the upper and lower boundaries of the current local search domain to generate an initial wandering point set with multi-scale characteristics, and construct the wandering trajectory step by step according to the priority of the mapping sequence scale.

[0143] S435. Utilize the historical difference between the antlion and the ant's position to calculate the dynamic trajectory update direction in real time, form a direction guiding factor, and combine it with the chaotic mapping sequence to generate an ant walking trajectory with direction preference.

[0144] S436. Record and extract the high-fitness segments of the walking trajectory in each iteration, and merge them in real time according to the degree of segment overlap to form candidate solutions of feature weight combination.

[0145] By dynamically adjusting the center and range of the local search domain, multi-scale initial position sequences are generated using multi-scale reconstructed chaotic mapping sequences to achieve fine local coverage. The trajectory update direction is guided by the position history difference, forming an ant walking trajectory with directional preference. High-fitness trajectory segments are merged in real time to form optimized feature weight combination candidate solutions, which effectively improves the search accuracy and convergence efficiency of the chaotic antlion optimization algorithm and enhances the fineness of abnormal pattern recognition parameter optimization.

[0146] In this embodiment, S5 specifically includes:

[0147] S51. Based on the optimal feature weight combination, perform feature weight weighting processing on each dimension of each transaction data in the financial compliance monitoring data set, and calculate and obtain the comprehensive abnormal feature score of each transaction data.

[0148] S52. Using the anomaly threshold, compare the comprehensive anomaly feature score of each transaction data with the numerical value of the anomaly threshold to determine and mark the anomaly status.

[0149] S53. For transaction data marked as abnormal, extract the account type classification label, transaction time window label, real-time dynamic direction label of fund flow, transaction amount value and corresponding comprehensive abnormal feature score in chronological order to form a structured abnormal transaction information record.

[0150] S54. Based on the account type classification tags and real-time dynamic direction tags of fund flow in the structured abnormal transaction information records, perform correlation path analysis on the data flow of abnormal transactions to determine the associated account paths of abnormal transactions.

[0151] S55. For each abnormal transaction's associated account path, analyze each path and determine the transaction risk level of each path based on the number of account nodes in the path, the transaction amount, and the comprehensive abnormal characteristic score.

[0152] S56. Integrate the structured information records of abnormal transactions, the associated account paths, and the determined transaction risk levels to form the final structured abnormal pattern recognition result.

[0153] By identifying abnormal transactions through feature weighting and anomaly threshold determination, structurally extracting transaction information to establish abnormal transaction records, deeply analyzing the paths of abnormal transactions related to accounts and refining the risk level for each path, and finally integrating them to form a structured abnormal pattern recognition result, the accuracy of abnormal transaction identification and the refinement of risk assessment are effectively improved, providing intuitive and reliable data support for the timely handling and decision-making of financial compliance risks.

[0154] In this embodiment, S6 specifically includes:

[0155] S61. Calculate the dynamically adjusted warning trigger priority based on the transaction risk level and the historical anomaly handling records of the transaction account in the structured anomaly pattern recognition results.

[0156] S62. By combining dynamically adjusted warning trigger priorities with predefined warning message generation templates, personalized warning message content is dynamically generated based on the differences in the characteristics of abnormal transaction account type classification tags, transaction time window tags, and real-time dynamic direction tags of fund flow.

[0157] S63. Based on the real-time status and historical response time records of the terminal devices of the designated compliance management personnel, select and adjust the priority level and push mechanism of the message push interface in real time.

[0158] S64. Utilize the priority matching results between warning messages and push interfaces to automatically and in real time generate enhanced standardized data messages containing abnormal transaction data feature information and suggested response strategies.

[0159] S65. Monitor the network communication quality of the terminal devices of designated compliance management personnel in real time during message push, dynamically adjust the segmentation strategy and transmission batch of push data packets, and optimize the success rate of warning message push.

[0160] S66. Based on the real-time response status and reading confirmation time of the terminal device, automatically update the push priority, push strategy and the response model of the terminal device for warning messages, and optimize the push process for the next warning message in real time.

[0161] By dynamically adjusting the warning trigger priority and generating dynamic warning messages using personalized templates, the message push strategy and priority are adaptively adjusted based on the real-time status and historical response performance of compliance management personnel's terminal devices. The network communication quality is monitored in real time to optimize the push success rate, and the push mechanism and response model are automatically updated using terminal feedback. This significantly improves the timeliness and reliability of warning messages, enhances the response efficiency of financial anomaly warnings, and improves the compliance management effect.

[0162] In this embodiment, S7 specifically includes:

[0163] S71. Acquire and record warning false alarms and omissions after manual confirmation by compliance management personnel, and store them in the data buffer according to the comprehensive abnormal feature score of transaction data, account type classification label, transaction time window label, and real-time dynamic direction label of fund flow.

[0164] S72. Based on the false alarms and missed alarms in the data buffer, analyze the deviation of each transaction data in the dimension of abnormal characteristics, and determine the error distribution in transaction frequency characteristics, transaction time sequence characteristics and network correlation characteristics.

[0165] S73. Based on the determined error distribution, calculate and generate the adjustment factor corresponding to the feature dimension. The adjustment factor is used to update the initial search space boundary of the chaotic antlion optimization algorithm.

[0166] S74. Based on the historical statistics and frequency of false alarms and omissions in transaction data, calculate the dynamic adjustment amount used to update the chaotic mapping control parameters in the chaotic antlion optimization algorithm.

[0167] S75. Update the upper and lower boundary values ​​of the initial search space of the chaotic antlion optimization algorithm in real time based on the adjustment factor, and synchronously input the dynamically adjusted chaotic mapping control parameters into the chaotic antlion optimization algorithm.

[0168] S76. Based on the predefined continuous evolution rules, automatically update the initial search space boundary values ​​and dynamically adjust the chaotic mapping control parameters.

[0169] By recording false alarms and missed alarms reported by compliance management personnel, the error distribution is analyzed to generate feature dimension adjustment factors and dynamically update the search space boundary and chaotic mapping control parameters of the chaotic antlion optimization algorithm. The algorithm parameter configuration is optimized in real time, thereby effectively improving the adaptability and accuracy of the algorithm to the identification of actual abnormal transactions and continuously improving the accuracy and stability of the system's anomaly detection model.

[0170] Example 1:

[0171] To verify the feasibility of this invention in practice, it was applied to a real-world scenario of financial compliance management in the financial center of a large enterprise group. This involved real-time monitoring and abnormal transaction early warning analysis of fund transfers across multiple subsidiaries within the group, covering a large volume of daily transactions. In this scenario, the company's existing financial compliance management methods relied primarily on manual review and fixed-rule early warnings. These methods suffered from delayed rule updates and an inability to respond in real-time to changes in abnormal patterns, making it difficult to identify numerous unknown or novel abnormal transaction patterns in a timely manner. Furthermore, due to the massive volume of financial data and the drastic changes in data flow density, traditional static rule-based methods could not adjust monitoring thresholds dynamically according to data changes. This resulted in a high false positive rate and frequent missed detections, impacting the efficiency and accuracy of financial compliance risk control.

[0172] In practice, the company deployed the financial compliance intelligent early warning system based on anomaly pattern mining proposed in this invention. First, a large amount of transaction data, including transaction amount, transaction time, transaction account, and fund flow details, is automatically acquired from the company's centralized financial management platform via a real-time data acquisition interface. This data is then encapsulated in a unified format to form a complete multi-dimensional data structure. After data preprocessing, a dynamic field constraint mechanism and adaptive interpolation algorithm automatically and accurately clean and impute missing, abnormal, and redundant data, achieving data structuring and high-quality standardization, laying the foundation for subsequent feature extraction.

[0173] Subsequently, the system continuously segments and analyzes standardized financial data using a dynamic sliding time window, automatically adjusting the window width and sliding step size to ensure real-time and accurate adaptation to changes in data flow density. Each sliding window's data is automatically constructed into a transaction network graph, with nodes representing trading accounts and edges between nodes representing fund flow relationships between accounts. Through analysis of node centrality and historical activity, the system extracts transaction frequency characteristics, transaction timing characteristics, and the number of effectively associated accounts for each node, forming a real-time updated anomaly pattern feature matrix.

[0174] During the feature optimization phase, the system's built-in chaotic antlion optimization algorithm, aiming to maximize detection accuracy and minimize false positive rate, underwent multiple rounds of iterative optimization. In the 40 iterations of actual operation, the algorithm used logistic chaotic mapping to determine the initial ant positions and automatically calculated the optimal feature weight combination and anomaly threshold through a dynamic adjustment mechanism of the local search domain and a chaotic random trajectory update strategy. The optimized model achieved an accuracy of 98.7% on the validation set of historical transaction data samples from enterprises, a significant improvement compared to the 85.4% accuracy of traditional static rule-based methods, while also reducing the false positive rate by approximately 70%.

[0175] Subsequently, the system utilizes the optimized feature weight combination and anomaly threshold to continuously analyze real-time transaction data and output structured identification results for abnormal transaction behavior in real time. For identified abnormal transactions, the system automatically and quickly determines the transaction risk level based on characteristics such as account type, transaction amount, and transaction time, and fully tracks and presents the abnormal transaction paths involved in a structured manner, enabling corporate compliance personnel to quickly and clearly grasp the details and associated risks of each abnormal transaction.

[0176] The system's intelligent dynamic early warning module automatically adjusts message priority and push strategies based on different risk levels and the historical handling of abnormal accounts. It generates and accurately pushes personalized abnormal transaction warning messages to compliance management personnel's terminals in real time. The real-time response and terminal feedback mechanism ensures the reliability of warning information transmission. Actual testing shows a 99.3% success rate for warning message pushes, and message response time has been reduced from an average of over one hour under traditional manual processing to less than two minutes on average.

[0177] This system also possesses continuous evolution capabilities. After processing each anomaly alert, enterprise compliance managers automatically upload manually confirmed false positives and false negatives to the system's data buffer. The system uses this false positive and false negative data to automatically analyze feature dimension deviations and update the parameter configuration and search space boundaries of the chaotic antlion algorithm in real time. During three months of continuous operation, the system underwent eight continuous evolutionary updates and iterations. The model's false positive rate decreased from 5.6% at the initial deployment to a final 1.2%, and the false negative rate also decreased from 7.3% to 1.5%, demonstrating significant continuous model optimization.

[0178] The table below shows a comparison of the accuracy of monitoring typical abnormal financial transactions before and after the deployment of the system of this invention:

[0179] Table 1 Comparison of Accuracy Data for Monitoring Abnormal Financial Transactions

[0180] Monitoring indicators Before system deployment (traditional method) After system deployment (method of this invention) Anomaly detection accuracy 85.4％ 98.7％ Average false alarm rate 7.8％ 1.2％ Average false negative rate 9.5％ 1.5％ Warning message response time Approximately 1 hour Within 2 minutes

[0181] The data analysis above clearly demonstrates that the system of this invention has achieved significant breakthroughs in anomaly identification accuracy, false alarm and false negative rates, and early warning message response speed. Statistical data three months after deployment shows that the frequency of manual intervention in corporate financial risk events has decreased by nearly 80%, with an average monthly saving of over 40% in manual review costs. Simultaneously, through precise tracing of anomaly paths and clear assessment of risk levels, the ability of financial compliance personnel to respond to risks has been greatly enhanced, significantly improving the efficiency of financial risk decision-making and the quality of financial management.

[0182] The practical application results of this embodiment fully verify the technical feasibility and practical effectiveness of the financial compliance intelligent early warning system proposed in this invention. In particular, it demonstrates significant technical advantages and application value in scenarios involving large-scale real-time transaction data analysis, complex dynamic feature optimization, and accurate and rapid identification of abnormal transactions.

[0183] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A financial compliance intelligent early warning system based on anomaly pattern mining, characterized in that, Includes the following modules: The real-time data acquisition module is used to acquire transaction data from the financial system in real time and generate a multi-dimensional data structure containing transaction amount, transaction time, account information, and related party information. The data preprocessing module is used to clean, impute missing values, and standardize multidimensional data structures to form a financial compliance monitoring data set. The abnormal pattern dynamic feature extraction module is used to transform the financial compliance monitoring data set into a network structure diagram composed of transaction sequences based on a dynamic sliding time window, and extract the temporal features, frequency features and network correlation features of transaction events to generate an abnormal pattern feature matrix. The feature optimization and selection module is used to input the abnormal pattern feature matrix into the chaotic antlion optimization algorithm, and then obtain the best feature weight combination and abnormal threshold for abnormal pattern recognition through the chaotic dynamic search mechanism of hunting and random walk. The real-time anomaly pattern identification module is used to analyze and monitor financial transaction data in real time by utilizing the optimal feature weight combination and anomaly threshold, and generate anomaly pattern identification results when it detects transaction behavior that deviates from the normal pattern. The intelligent dynamic early warning module is used to automatically trigger and push warning information to compliance management personnel in real time after receiving the abnormal pattern recognition results; The model continuous evolution module is used to receive and store warning false alarms and missed alarms after manual confirmation, and update the search space and parameter configuration of the chaotic antlion optimization algorithm.

2. The financial compliance intelligent early warning system based on anomaly pattern mining according to claim 1, characterized in that, The modules are connected in the following way: S1. Obtain transaction data in real time from the financial system interface and generate a multi-dimensional data structure in a unified format; S2. Through data cleaning rules and data standardization algorithms, multi-dimensional data structures are processed in a unified manner to form a set of financial compliance monitoring data. S3. Based on the preset dynamic sliding time window rules, the financial compliance monitoring data set is sequentially transformed into a serialized network structure diagram, and the transaction event time sequence features, frequency features and network correlation features are extracted from the network structure diagram node by node, and the abnormal pattern feature matrix is ​​output. S4. Call the chaotic antlion optimization algorithm built into the feature optimization selection module. Through the chaotic dynamic search mechanism that combines hunting position selection and random walk trajectory optimization, automatically calculate and determine the best feature weight combination and abnormal threshold in the abnormal pattern feature matrix for abnormal pattern recognition. S5. Utilize the optimal feature weight combination and anomaly threshold to continuously analyze and monitor financial transaction data in the financial compliance monitoring dataset. When transaction data is found to deviate significantly from the normal pattern, output the structured anomaly pattern identification result. S6. Based on predefined warning rules, warning messages are automatically generated in real time according to the structured anomaly pattern recognition results, and the warning push interface is called to send them to the terminal devices of designated compliance management personnel in an instant. S7. Based on the warning false alarms and omissions confirmed by compliance management personnel, the data is automatically stored in the data buffer, and the search space and parameter configuration of the chaotic antlion optimization algorithm are dynamically updated according to the predefined continuous evolution rules.

3. The financial compliance intelligent early warning system based on anomaly pattern mining according to claim 2, characterized in that, S1 specifically includes: S11. Construct a dual-channel asynchronous data acquisition channel between the financial system interface and the intelligent early warning system. Channel 1 adopts a periodic active polling mode, while Channel 2 adopts a real-time triggering mode based on transaction events. The two channels run in parallel and independently, and each is set with an independent acquisition clock. S12. Embed a unique transaction sequence tag and a channel source tag into each transaction data acquired asynchronously through dual channels, and establish and store a one-to-one mapping relationship between the tag and the transaction data. S13. Real-time call to the asynchronous data cross-validation mechanism, through the mapping relationship of the identification tags, to compare and verify the data of the same transaction event obtained by Channel 1 and Channel 2 respectively, automatically identify and mark the existence of field data differences, data missing and duplicate reception; S14. Based on the cross-validation mechanism, automatically generate a data consistency verification report, and use data compensation and deduplication rules to compensate for the difference of missing data, intelligently merge duplicate data, and dynamically and adaptively repair transaction data. S15. Based on the automatically repaired transaction data, construct a three-dimensional transaction attribute label set including account type classification labels, transaction time window labels, and real-time dynamic direction labels of fund flow, and perform labeling processing on the transaction data. S16. After the transaction data has been processed by three-dimensional transaction attribute tagging, it is encapsulated one by one according to the unified data structure standard rules to generate a multi-dimensional data structure.

4. The financial compliance intelligent early warning system based on anomaly pattern mining according to claim 2, characterized in that, S2 specifically includes: S21. A dynamic field constraint mechanism is adopted to detect the changing trend of transaction amount values ​​and account classification labels in multidimensional data structures in real time, and to automatically update the outlier identification threshold to identify data with abnormal field values. S22. Construct a multi-dimensional data structure association graph based on the time window label of the trading period and the real-time dynamic direction label of the fund flow, and identify data units with missing or incomplete association relationships; S23. By comparing the topological position of the multidimensional data structure association graph with the identified abnormal data units in context, extract data reference values ​​from the upstream and downstream data units adjacent to the topological position, and dynamically correct the abnormal data units. S24. Based on the historical distribution of account type classification labels and transaction amount data, generate a personalized data integrity model, detect and automatically fill in data units with missing or incomplete relationships; S25. By constructing a dynamic feature space for transaction data, adaptive nonlinear mapping is performed on the corrected and filled transaction data to generate standardized data with uniformly distributed data features. S26. Utilize the time-series data balanced distribution rules to verify the time-series continuity and integrity of standardized data, and generate a financial compliance monitoring data set.

5. A financial compliance intelligent early warning system based on anomaly pattern mining according to claim 2, characterized in that, S3 specifically includes: S31. Based on the statistical indicators of data density and transaction frequency of each data unit in the financial compliance monitoring data set, calculate and adjust the window width and sliding step of the dynamic sliding time window to form a non-fixed length sliding time window that dynamically adapts to the data flow density. S32. For each set of financial compliance monitoring data within a non-fixed length sliding time window, analyze each transaction data item by item in the order of the occurrence of the transaction events, determine the relationship between the transaction entity node and the transaction, and establish a dynamic time-series network structure diagram of the transaction events. S33. Based on the positional centrality and historical transaction activity of each node in the dynamic time-series network structure graph, dynamically calculate the node transaction frequency weighting coefficient, and use the node transaction frequency weighting coefficient to adjust the original transaction frequency of the node node by node to generate the node activity weighted transaction frequency feature. S34. For each node, record and extract the occurrence time of node transaction events in the dynamic temporal network structure diagram, calculate the time interval difference between adjacent transaction events, and form a node transaction temporal feature sequence that is sensitive to transaction intervals. S35. Based on the dual constraints of the order of transaction occurrence and the difference in transaction amount between nodes, evaluate and eliminate invalid transaction associations, retain only valid node connections with temporal causal relationship and amount matching relationship, calculate the number of connections between a node and all valid associated nodes, and generate node valid transaction network association characteristics. S36. Based on the dynamic changes in the transaction frequency characteristics, transaction timing characteristics, and network association characteristics of nodes within the current window data, calculate the dynamic fusion weight coefficients of various features. Through the adaptive combination rules of the dynamic fusion weight coefficients, fuse all features of the nodes to construct an abnormal pattern feature matrix with dynamic feature weight adjustment.

6. The financial compliance intelligent early warning system based on anomaly pattern mining according to claim 2, characterized in that, S4 specifically includes: S41. Set the initial search space for the chaotic antlion optimization algorithm, and use the transaction frequency feature, transaction time sequence feature and network correlation feature in the abnormal mode feature matrix as the feature dimensions to be optimized. S42. According to the logistic chaotic mapping equation, initialize the initial position of the ant in the algorithm; S43. In each iteration, a local search domain is constructed based on the current hunting position of the antlion. The ant's walking trajectory in the local search domain is randomly generated based on the chaotic mapping. The walking trajectory is defined as a candidate solution of the feature weight combination. S44. The classification error rate of the abnormal mode feature matrix is ​​used as the fitness function to evaluate the fitness of candidate solutions in real time. S45. Record the fitness function evaluation results, select the feature weight combination with the highest fitness and the corresponding anomaly threshold as the optimal candidate solution for the current iteration, and dynamically adjust the antlion hunting position and local search domain according to the evaluation results. S46. Set the algorithm iteration termination condition as the change in the fitness function over M consecutive iterations being less than the convergence threshold ∈. After reaching the termination condition, output the final determined optimal feature weight combination and anomaly threshold.

7. The financial compliance intelligent early warning system based on anomaly pattern mining according to claim 6, characterized in that, Specifically, S43 includes: S431. Based on the current hunting position of the antlion and the changing trend of the hunting position in the historical iteration, calculate and dynamically adjust the center position of the local search domain, and determine the center of the local search domain by the weighted historical average value of the position change vector. S432. Based on the concentration of the global optimal position on the feature dimension, adjust the range coefficient of the local search domain in real time, and dynamically expand and shrink the local search domain. S433. Based on the mapping sequence of the logistic chaotic mapping equation, perform multi-scale reconstruction, perform multi-scale hierarchical sampling on the mapping sequence, generate chaotic initial position sequences at different scales, and perform multi-scale coverage on the local search domain. S434. Merge the multi-scale chaotic initial position sequence with the upper and lower boundaries of the current local search domain to generate an initial wandering point set with multi-scale characteristics, and construct the wandering trajectory step by step according to the priority of the mapping sequence scale. S435. Utilize the historical difference between the antlion and the ant's position to calculate the dynamic trajectory update direction in real time, form a direction guiding factor, and combine it with the chaotic mapping sequence to generate an ant walking trajectory with direction preference. S436. Record and extract the high-fitness segments of the walking trajectory in each iteration, and merge them in real time according to the degree of segment overlap to form candidate solutions of feature weight combination.

8. The financial compliance intelligent early warning system based on anomaly pattern mining according to claim 1, characterized in that, S5 specifically includes: S51. Based on the optimal feature weight combination, perform feature weight weighting processing on each dimension of each transaction data in the financial compliance monitoring data set, and calculate and obtain the comprehensive abnormal feature score of each transaction data. S52. Using the anomaly threshold, compare the comprehensive anomaly feature score of each transaction data with the numerical value of the anomaly threshold to determine and mark the anomaly status. S53. For transaction data marked as abnormal, extract the account type classification label, transaction time window label, real-time dynamic direction label of fund flow, transaction amount value and corresponding comprehensive abnormal feature score in chronological order to form a structured abnormal transaction information record. S54. Based on the account type classification tags and real-time dynamic direction tags of fund flow in the structured abnormal transaction information records, perform correlation path analysis on the data flow of abnormal transactions to determine the associated account paths of abnormal transactions. S55. For each abnormal transaction's associated account path, analyze each path and determine the transaction risk level of each path based on the number of account nodes in the path, the transaction amount, and the comprehensive abnormal characteristic score. S56. Integrate the structured information records of abnormal transactions, the associated account paths, and the determined transaction risk levels to form the final structured abnormal pattern recognition result.

9. A financial compliance intelligent early warning system based on anomaly pattern mining as described in claim 1, characterized in that, S6 specifically includes: S61. Calculate the dynamically adjusted warning trigger priority based on the transaction risk level and the historical anomaly handling records of the transaction account in the structured anomaly pattern recognition results. S62. By combining dynamically adjusted warning trigger priorities with predefined warning message generation templates, personalized warning message content is dynamically generated based on the differences in the characteristics of abnormal transaction account type classification tags, transaction time window tags, and real-time dynamic direction tags of fund flow. S63. Based on the real-time status and historical response time records of the terminal devices of the designated compliance management personnel, select and adjust the priority level and push mechanism of the message push interface in real time. S64. Utilize the priority matching results between warning messages and push interfaces to automatically and in real time generate enhanced standardized data messages containing abnormal transaction data feature information and suggested response strategies. S65. Monitor the network communication quality of the terminal devices of designated compliance management personnel in real time during message push, dynamically adjust the segmentation strategy and transmission batch of push data packets, and optimize the success rate of warning message push. S66. Based on the real-time response status and reading confirmation time of the terminal device, automatically update the push priority, push strategy and the response model of the terminal device for warning messages, and optimize the push process for the next warning message in real time.

10. A financial compliance intelligent early warning system based on anomaly pattern mining according to claim 1, characterized in that, Specifically, S7 includes: S71. Acquire and record warning false alarms and omissions after manual confirmation by compliance management personnel, and store them in the data buffer according to the comprehensive abnormal feature score of transaction data, account type classification label, transaction time window label, and real-time dynamic direction label of fund flow. S72. Based on the false alarms and missed alarms in the data buffer, analyze the deviation of each transaction data in the dimension of abnormal characteristics, and determine the error distribution in transaction frequency characteristics, transaction time sequence characteristics and network correlation characteristics. S73. Based on the determined error distribution, calculate and generate the adjustment factor corresponding to the feature dimension. The adjustment factor is used to update the initial search space boundary of the chaotic antlion optimization algorithm. S74. Based on the historical statistics and frequency of false alarms and omissions in transaction data, calculate the dynamic adjustment amount used to update the chaotic mapping control parameters in the chaotic antlion optimization algorithm. S75. Update the upper and lower boundary values ​​of the initial search space of the chaotic antlion optimization algorithm in real time based on the adjustment factor, and synchronously input the dynamically adjusted chaotic mapping control parameters into the chaotic antlion optimization algorithm. S76. Based on the predefined continuous evolution rules, automatically update the initial search space boundary values ​​and dynamically adjust the chaotic mapping control parameters.

Images