A financial information security protection method and system based on artificial intelligence

Abstract

The present invention relates to the field of information security protection technology, and in particular to a financial information security protection method and system based on artificial intelligence. The method comprises the following steps: acquiring and identifying multimodal financial information, and grouping it into financial transaction behavior information blocks and user basic information blocks; analyzing and trend-predicting transaction behavior information to generate predicted behavior data; hierarchically correlating and mapping user basic information based on predicted data to obtain user behavior consensus data; dynamically jointly encrypting multimodal financial information using these consensus data to generate preliminary encrypted information; simulating vulnerability attacks and screening security threats to obtain security threat data; and re-encrypting encrypted information based on threat data to generate more secure re-encrypted information. The present invention achieves efficient and secure financial information protection.

Description

Technical Field

[0001] The present invention relates to the field of information security protection technology, and in particular to a financial information security protection method and system based on artificial intelligence. Background Art

[0002] With the rapid development of financial technology, financial institutions have gradually relied on multimodal data for decision-making and operations. These data include text, voice, images, user personal information, transaction records and other forms of information. While these multimodal financial information bring business innovation, they also trigger new security challenges. Traditional encryption and protection technologies are powerless in dealing with complex and diverse financial information security threats, especially in the face of increasingly complex network attacks. Existing technologies are difficult to provide adequate protection. The current financial information security system mainly relies on static encryption algorithms and predefined security policies. These methods often cannot be adjusted in time when facing unknown or variant attacks, resulting in security loopholes. In addition, with the continuous evolution of attack technology, attackers have begun to use advanced attack strategies, such as using artificial intelligence to generate adversarial samples, thereby bypassing traditional security protection mechanisms. Existing financial information security protection is not flexible and intelligent enough to face the ever-changing threat environment, resulting in financial information being attacked. Summary of the invention

[0003] Based on this, it is necessary to provide an artificial intelligence-based financial information security protection method and system to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above purpose, a financial information security protection method based on artificial intelligence includes the following steps:

[0005] Step S1: Acquire multimodal financial information; identify the modal type of the multimodal financial information and group the information blocks to obtain a financial transaction behavior information block and a user basic information block;

[0006] Step S2: Analyze the behavior pattern of the financial transaction behavior information block, predict the behavior trend, and generate predicted behavior data;

[0007] Step S3: Based on the predicted behavior data, hierarchical association mapping is performed on the user basic information blocks to obtain user behavior consensus data;

[0008] Step S4: dynamically jointly encrypt multimodal financial information using user behavior consensus data to generate encrypted financial information;

[0009] Step S5: simulating vulnerability attacks on encrypted financial information based on the predicted behavior data, and performing security threat screening to obtain security threat data;

[0010] Step S6: re-encrypt the encrypted financial information based on the security threat data to generate re-encrypted financial information.

[0011] The present invention effectively classifies and structures complex multi-dimensional data by performing modal type identification and information block grouping on multi-modal financial information, thereby improving the efficiency and accuracy of data analysis and processing. By performing pattern analysis and trend prediction on financial transaction behaviors, it is possible to identify potential risk behaviors in advance and generate predicted behavior data, which is helpful to take protective measures in advance. Dynamic joint encryption is performed using user behavior consensus data, so that encryption strategies can be adjusted according to real-time behavior data, thereby enhancing the flexibility and pertinence of encryption and improving the security of information. Vulnerability attack simulation and security threat screening based on predicted behavior data can effectively identify potential security threats and provide timely risk warnings and protection. Secondary encryption based on security threat data after initial encryption further strengthens the security protection of financial information, prevents various types of attacks, and ensures the integrity and confidentiality of information.

[0012] The present invention also provides an artificial intelligence-based financial information security protection system, which is used to execute the artificial intelligence-based financial information security protection method as described above. The artificial intelligence-based financial information security protection system includes:

[0013] The information acquisition module is used to acquire multimodal financial information; identify the modal type of the multimodal financial information, and group the information blocks to obtain the financial transaction behavior information block and the user basic information block;

[0014] The behavior prediction module is used to analyze the behavior patterns of the financial transaction behavior information blocks, predict the behavior trends, and generate predicted behavior data;

[0015] The information mapping module is used to perform hierarchical association mapping on the user basic information blocks based on the predicted behavior data to obtain the user behavior consensus data;

[0016] Joint encryption module, used to dynamically joint encrypt multimodal financial information using user behavior consensus data to generate encrypted financial information;

[0017] A threat screening module is used to simulate vulnerability attacks on encrypted financial information based on predicted behavior data, and to perform security threat screening to obtain security threat data;

[0018] The secondary encryption module is used to perform secondary encryption on encrypted financial information based on security threat data to generate re-encrypted financial information.

[0019] The present invention can identify and classify different types of multimodal financial information by utilizing the information acquisition module, divide complex financial data into behavioral information blocks and basic information blocks, and enhance the accuracy and efficiency of the system in processing multidimensional data. The behavioral prediction module analyzes the financial transaction behavior pattern, performs trend prediction, and identifies potential risks and abnormal behaviors in advance, thereby improving the system's foresight and responsiveness. The information mapping module uses the predicted behavior data to perform hierarchical association mapping and generates user behavior consensus data, so that the association between user basic information and behavior information is closer and more accurate, providing a solid foundation for subsequent encryption and protection. The joint encryption module uses dynamic joint encryption technology to adjust the encryption strategy in real time according to the user behavior consensus data, making the encryption process more flexible and able to effectively respond to changing security threats. The threat screening module simulates vulnerability attacks and performs security threat screening to timely discover potential threats in encrypted information, which helps to identify and prevent security vulnerabilities in advance and enhance the overall protection capability of the system. The secondary encryption module performs secondary encryption in combination with security threat data on the basis of the initial encryption, further improving the confidentiality and security of financial information, and ensuring that high-intensity protection can be maintained in the face of complex attacks. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of the steps of a financial information security protection method based on artificial intelligence;

[0021] Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart;

[0022] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0023] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0024] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0025] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0026] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0027] To achieve this, please refer to Figures 1 to 3 , a financial information security protection method based on artificial intelligence, comprising the following steps:

[0028] Step S1: Acquire multimodal financial information; identify the modal type of the multimodal financial information and group the information blocks to obtain a financial transaction behavior information block and a user basic information block;

[0029] Step S2: Analyze the behavior pattern of the financial transaction behavior information block, predict the behavior trend, and generate predicted behavior data;

[0030] Step S3: Based on the predicted behavior data, hierarchical association mapping is performed on the user basic information blocks to obtain user behavior consensus data;

[0031] Step S4: dynamically jointly encrypt multimodal financial information using user behavior consensus data to generate encrypted financial information;

[0032] Step S5: simulating vulnerability attacks on encrypted financial information based on the predicted behavior data, and performing security threat screening to obtain security threat data;

[0033] Step S6: re-encrypt the encrypted financial information based on the security threat data to generate re-encrypted financial information.

[0034] The present invention effectively classifies and structures complex multi-dimensional data by performing modal type identification and information block grouping on multi-modal financial information, thereby improving the efficiency and accuracy of data analysis and processing. By performing pattern analysis and trend prediction on financial transaction behaviors, it is possible to identify potential risk behaviors in advance and generate predicted behavior data, which is helpful to take protective measures in advance. Dynamic joint encryption is performed using user behavior consensus data, so that encryption strategies can be adjusted according to real-time behavior data, thereby enhancing the flexibility and pertinence of encryption and improving the security of information. Vulnerability attack simulation and security threat screening based on predicted behavior data can effectively identify potential security threats and provide timely risk warnings and protection. Secondary encryption based on security threat data after initial encryption further strengthens the security protection of financial information, prevents various types of attacks, and ensures the integrity and confidentiality of information.

[0035] In the embodiment of the present invention, refer to Figure 1 , is a schematic diagram of the steps of a financial information security protection method based on artificial intelligence of the present invention. In this example, the financial information security protection method based on artificial intelligence includes the following steps:

[0036] Step S1: Acquire multimodal financial information; identify the modal type of the multimodal financial information and group the information blocks to obtain a financial transaction behavior information block and a user basic information block;

[0037] In this embodiment, multimodal financial information is obtained, and modal type identification and information block grouping are performed. Financial information of different modalities is collected through multiple data collection terminals or systems deployed in financial institutions, including but not limited to text data, image data, audio data, and video data. Convolutional neural networks (CNNs) are used to identify the modal type of multimodal data and determine the specific category of each data, such as classifying transaction records as text data and classifying customer identity data as image data. Through a grouping method based on a clustering algorithm, information of the same modality type is integrated into corresponding information blocks, thereby forming two major data sets: financial transaction behavior information blocks and user basic information blocks.

[0038] Step S2: Analyze the behavior pattern of the financial transaction behavior information block, predict the behavior trend, and generate predicted behavior data;

[0039] In this embodiment, the behavior pattern analysis is performed on the financial transaction behavior information block, and the behavior trend prediction is performed to generate predicted behavior data. First, the historical transaction data in the financial transaction behavior information block is modeled using a time series analysis algorithm (such as an LSTM network) to capture the behavior patterns therein, including common transaction frequencies, transaction amounts, and transaction time period characteristics. Based on the identified patterns, a prediction model such as a GRU network is used to predict the behavior trend and output future transaction behavior trends. The trend data includes not only conventional transaction activity predictions, but also potential abnormal transaction behaviors, such as large-scale fund transfers or multiple cross-border transactions, thereby generating accurate predicted behavior data.

[0040] Step S3: Based on the predicted behavior data, hierarchical association mapping is performed on the user basic information blocks to obtain user behavior consensus data;

[0041] In this embodiment, the user basic information block is hierarchically associated and mapped based on the predicted behavior data to obtain the user behavior consensus data. The predicted behavior data and the user basic information block are multi-dimensionally associated and mapped using the association analysis technology based on the graph database. Specifically, it includes constructing user information nodes and behavior data nodes, and establishing association edges between nodes through key identifiers such as user ID and transaction account. Then, a deep learning algorithm (such as the convolutional network, GCN) is used to hierarchically process the mapped data, identify the implicit association between the user and his behavior pattern, and generate a user-centered behavior consensus data set. It reflects the overall behavioral characteristics of each user in a specific predicted behavior context.

[0042] Step S4: dynamically jointly encrypt multimodal financial information using user behavior consensus data to generate encrypted financial information;

[0043] In this embodiment, user behavior consensus data is used to dynamically jointly encrypt multimodal financial information to generate encrypted financial information. User-specific encryption key parameters are extracted based on user behavior consensus data, and these parameters are closely related to the user's behavior pattern. A distributed key generation algorithm (such as the Shamir secret sharing scheme) is used to dynamically disperse the encryption key of multimodal financial information into each user behavior consensus data. The entire multimodal financial information block is jointly encrypted by a symmetric encryption algorithm (such as AES-256) combined with the behavior consensus key parameters to ensure high security of data in storage and transmission, and finally generate encrypted financial information data.

[0044] Step S5: simulating vulnerability attacks on encrypted financial information based on the predicted behavior data, and performing security threat screening to obtain security threat data;

[0045] In this embodiment, vulnerability attack simulation is performed on encrypted financial information based on predicted behavior data, and security threat screening is performed to obtain security threat data. Generative adversarial networks (GAN) are used to simulate potential attack patterns, and different types of attack scenarios are constructed through adversarial sample generation technology, such as simulated man-in-the-middle attacks, side channel attacks, etc., to perform vulnerability testing on encrypted financial information. Deep packet inspection technology (DPI) is used to comprehensively screen the generated attack data, identify potential security threats and vulnerabilities, such as weaknesses in encryption algorithms or risks in key management, and generate security threat data. After further analysis, it will be used to guide subsequent protective measures.

[0046] Step S6: re-encrypt the encrypted financial information based on the security threat data to generate re-encrypted financial information.

[0047] In this embodiment, encrypted financial information is encrypted twice based on security threat data to generate re-encrypted financial information. The security threat data generated in the previous step is classified and weighted to identify the most serious security risks. For these high-risk areas, an encryption algorithm based on lattice theory (such as the NTRU encryption algorithm) is applied to perform a second encryption process on the previously encrypted financial information. Through the two encryption processes, the security of financial information is further improved, especially in high-risk threat areas, ensuring that the data can maintain confidentiality and integrity even when subjected to high-level attacks, and ultimately generating more secure re-encrypted financial information.

[0048] Preferably, step S1 comprises:

[0049] S11: acquiring multimodal financial information; performing feature extraction on the multimodal financial information to obtain a multidimensional feature vector;

[0050] Step S12: performing modal clustering analysis on the multimodal financial information according to the multidimensional feature vector to generate a modal clustering result;

[0051] Step S13: performing feature recognition on the modal clustering results class by class to obtain a class modal feature vector;

[0052] Step S14: performing modal type identification on the category modal feature vector to generate modal classification data;

[0053] Step S15: Group the multimodal financial information into information blocks based on the modal classification data to obtain financial transaction behavior information blocks and user basic information blocks.

[0054] The present invention can effectively convert different forms of data (such as text, images, sounds, etc.) into structured multi-dimensional feature vectors by acquiring multi-modal financial information and performing feature extraction, which helps to improve the comprehensive understanding and analysis capabilities of financial information. By using multi-dimensional feature vectors for modal clustering analysis, data with similar features can be automatically classified into the same category, making information processing more organized, thereby improving the efficiency and accuracy of data processing. By performing feature recognition on the modal clustering results, the modal feature vector of each category can be extracted, which helps to deeply understand the characteristics of each modality, thereby improving the detection and classification capabilities of specific patterns. By performing modal type recognition on the category modal feature vectors, it can It can generate more accurate modal classification data, which helps to better identify and distinguish different types of modal information, thereby improving the security protection level of financial information. By grouping multimodal financial information into information blocks based on modal classification data, the information can be divided into financial transaction behavior information blocks and user basic information blocks, and the relevant information can be organized in an orderly manner to support more effective analysis and decision-making. Comprehensive and in-depth analysis and classification of financial information helps to identify potential security threats and abnormal behaviors, thereby improving the security protection accuracy of financial information systems. Accurate modal clustering, feature recognition and classification capabilities help reduce false alarms and missed alarms, and ensure the effectiveness and reliability of financial information security protection measures.

[0055] In this embodiment, multimodal financial information is obtained, and features are extracted to obtain a multidimensional feature vector. Through the data acquisition module configured in the financial system, multimodal financial information including text (such as transaction records), images (such as customer ID card photos), audio (such as customer call records) and video (such as surveillance camera images) is obtained in real time. Feature extraction technology in deep learning, such as convolutional neural network (CNN), is used to process image data and extract image features. Word embedding technology (such as Word2Vec or BERT) is used to convert text data into numerical features; voiceprint recognition model is used to extract audio features from audio data, and temporal convolutional network (TCN) is used to extract temporal features from video data. All extracted features are concatenated to form a multi-dimensional feature vector, which covers the information features of each mode. Based on the multi-dimensional feature vector, the multi-modal financial information is subjected to modal clustering analysis to generate modal clustering results. The multi-dimensional feature vector is analyzed by a clustering method based on the K-means algorithm. The number of cluster centers is determined by selecting an appropriate K value. The feature vector is subjected to K-means clustering to group multi-modal financial information with similar features. Each cluster represents a clustering result, which contains financial data with similar features. For each cluster cluster, the mean vector of the data points in the cluster is calculated, and the modal clustering results are generated based on these mean vectors for further analysis and processing. The modal clustering results are subjected to class-by-class feature recognition to obtain the class modal feature vector. The feature recognition technology is applied to each modal clustering result. The clustering results are subjected to class-by-class feature recognition using a support vector machine (SVM) or a random forest classifier to extract the main features of each cluster and form a class feature vector. Each category feature vector describes the core features of this type of financial information. For example, financial transaction information includes features such as transaction time and amount, and user basic information includes user identity recognition data. Through the above category feature vectors, the features of each cluster category can be identified and described in detail, the category modal feature vectors can be identified as modal types, and modal classification data can be generated. The category modal feature vectors can be further processed using classification algorithms (such as the fully connected layer in a deep neural network). First, each category modal feature vector is input into the trained classification model for modal type identification.Through the label information in the training data, the model can map the category feature vector to a specific modal type, such as financial transaction behavior, basic user information, etc., to generate modal classification data. Each data point corresponds to its specific modal type identifier. Based on the modal classification data, the multimodal financial information is grouped into information blocks to obtain financial transaction behavior information blocks and user basic information blocks. The modal classification data is used to group the original multimodal financial information. By matching the modal type identifier in the modal classification data, the relevant information is integrated into two main information blocks: the financial transaction behavior information block and the user basic information block. The financial transaction behavior information block includes data related to transaction activities, such as transaction records, amounts, time, etc. The user basic information block includes data related to the user identity, such as user profile and ID card information.

[0056] Preferably, step S2 comprises:

[0057] Step S21: Decomposing the financial transaction behavior information block into time series features to obtain a time series feature sequence;

[0058] Step S22: performing subsequence pattern matching on the financial transaction behavior information block according to the time series feature sequence to generate a behavior pattern matching set;

[0059] Step S23: performing center point clustering processing on the behavior pattern matching set to obtain pattern center data;

[0060] Step S24: Analyze the change trend of the pattern center data to generate pattern trend data;

[0061] Step S25: Predicting the behavior trend of the financial transaction behavior information block based on the pattern trend data to generate predicted behavior data.

[0062] The present invention can extract detailed time series feature sequences by decomposing the financial transaction behavior information block, reveal the time dependency in the financial transaction behavior, and lay the foundation for subsequent behavior analysis. The time series feature sequences are used to perform subsequence pattern matching, and a behavior pattern matching set can be generated, which can identify similar transaction patterns and help identify common behavior patterns or potential abnormal behaviors. The behavior pattern matching set is subjected to center point clustering processing, and pattern center data can be obtained, which is helpful to summarize and summarize the main behavior patterns, thereby improving the overall understanding and analysis ability of the transaction behavior. The change trend analysis of the pattern center data can generate pattern trend data, which can reveal the law of the behavior pattern changing over time, and help identify long-term trends and abnormal fluctuations. Based on the pattern trend data, the behavior trend prediction of the financial transaction behavior information block can generate predicted behavior data, which can predict the future financial transaction behavior and provide forward-looking risk management and decision support. Through time series feature decomposition and behavior trend prediction, real-time monitoring and risk detection of financial transactions can be achieved, which helps to timely discover and respond to potential financial risks and security threats. Combining pattern matching and trend prediction can more accurately identify abnormal transaction patterns and potential fraudulent behavior, thereby improving the security protection level of financial information systems. Automated time series feature extraction, pattern matching and trend prediction technology can reduce manual intervention and operational burden, and improve the efficiency and reliability of financial information security protection systems.

[0063] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0064] Step S21: Decomposing the financial transaction behavior information block into time series features to obtain a time series feature sequence;

[0065] In this embodiment, the time series feature decomposition is performed on the financial transaction behavior information block to obtain a time series feature sequence, and each transaction data in the financial transaction behavior information block is processed into time series data. First, the transaction records are sorted by timestamp. The long short-term memory network (LSTM) is used to perform feature decomposition on these time series data to extract important features in the time series, such as transaction frequency, transaction amount fluctuation, etc. The transaction data is first normalized and then input into the LSTM model. The time series is feature extracted through multiple hidden layers of the model to generate a series of time series feature sequences.

[0066] Step S22: performing subsequence pattern matching on the financial transaction behavior information block according to the time series feature sequence to generate a behavior pattern matching set;

[0067] In this embodiment, the dynamic time warping (DTW) algorithm is used to perform subsequence pattern matching on the extracted time series feature sequence, the time series feature sequence is divided into multiple subsequences, and these subsequences are compared with known transaction patterns. The similarity between the subsequences and the template pattern is calculated by the DTW algorithm to identify the matching transaction behavior pattern, and all matching results are summarized into a behavior pattern matching set, which includes the identified transaction patterns and their corresponding matching degrees, providing data support for subsequent pattern analysis.

[0068] Step S23: performing center point clustering processing on the behavior pattern matching set to obtain pattern center data;

[0069] In this embodiment, the K-means clustering algorithm is applied to perform cluster analysis on the data in the behavior pattern matching set. According to the behavior pattern feature vector in the matching result, the K-means algorithm is used to set the number of cluster centers K, and the data is iteratively processed. The algorithm calculates the distance from each data point to each cluster center, and classifies the data points into the cluster with the nearest center point. Finally, the pattern center data is generated by calculating the coordinates of the center point of each cluster.

[0070] Step S24: Analyze the change trend of the pattern center data to generate pattern trend data;

[0071] In this embodiment, the generated pattern center data is subjected to a change trend analysis, and a time series analysis method, such as an autoregressive moving average (ARMA) model, is used to model the time change trend of the pattern center data. By smoothing and trend decomposing the pattern center data, long-term trends, cyclical changes, and random fluctuations are identified. These analysis results are integrated to generate pattern trend data, which shows the change trends of different financial behavior patterns over time and helps identify potential trend changes.

[0072] Step S25: Predicting the behavior trend of the financial transaction behavior information block based on the pattern trend data to generate predicted behavior data.

[0073] In this embodiment, a prediction model (such as a long short-term memory network LSTM in a recursive neural network) is applied to analyze the pattern trend data, the pattern trend data is input into the LSTM model, and the model is trained to identify the time series features and trends in the data. Through the training process of the model, future behavior trend prediction results are generated, and the future behavior trend prediction results include the changing trend of the financial transaction behavior pattern that will appear in the future, the expected transaction volume, etc., and finally the predicted behavior data is generated.

[0074] Preferably, step S3 comprises:

[0075] Step S31: performing continuous coherence analysis on the predicted behavior data to obtain a behavior topology feature spectrum;

[0076] Step S32: construct a multi-layer heterogeneous network for the user basic information block according to the behavior topology feature spectrum to obtain a user-behavior association network;

[0077] Step S33: performing adjacency matrix analysis on the user-behavior association network to generate a consensus information matrix;

[0078] Step S34: Perform hierarchical fusion processing on the consensus information matrix to obtain user behavior consensus data.

[0079] !By continuously coherently analyzing the predicted behavior data, we can obtain the behavior topology feature spectrum, which can reveal the complex structure and pattern of the behavior data, and help to more accurately understand the user's behavior characteristics and dynamic changes. Based on the behavior topology feature spectrum, the user basic information block is constructed with multiple layers of heterogeneous networks, which can generate a user-behavior association network, integrate user information and behavior data at different levels, and provide a comprehensive perspective for comprehensive analysis. Adjacency matrix analysis of the user-behavior association network can generate a consensus information matrix, which helps to reveal the correlation and consensus between user behaviors, thereby improving the understanding and recognition of user behavior patterns. Layered fusion processing of the consensus information matrix can obtain user behavior consensus data, which helps to integrate information at different levels, thereby generating more accurate and reliable user behavior consensus data. The construction of the behavior topology feature spectrum and the user-behavior association network, combined with the consensus information matrix and consensus data, can provide in-depth insights into user behavior, help identify users' potential behavior patterns and abnormal behaviors, and improve the accuracy of security protection. By analyzing the consensus data of user behavior, abnormal behavior can be detected and warned more effectively. For example, by discovering the inconsistency between user behavior and consensus data, fraud or security risks can be discovered early. The technical features of multi-layer heterogeneous network construction and hierarchical fusion processing can integrate data from different sources, enhance the comprehensive utilization value of data, and help to achieve comprehensive analysis and processing of financial information. Comprehensive analysis of user behavior consensus data and network structure can help optimize risk management strategies and enhance the security protection capabilities of financial information systems.

[0080] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0081] Step S31: performing continuous coherence analysis on the predicted behavior data to obtain a behavior topology feature spectrum;

[0082] In this embodiment, data preprocessing is performed on the time series in the predicted behavior data, the data range is scaled to a uniform interval using a standardized method, and the processed data is analyzed using a persistent homology technique. The persistent homology technique uses generated simplified shapes (such as point sets, boundaries, and holes) to capture the topological features of the data. The Betti number of the data is calculated using a persistent homology algorithm to obtain the persistence of each topological feature, thereby generating a behavioral topological feature spectrum. The feature spectrum represents the persistence of each topological structure in the data and can reveal the long-term structure and change pattern in the predicted behavior data.

[0083] Step S32: construct a multi-layer heterogeneous network for the user basic information block according to the behavior topology feature spectrum to obtain a user-behavior association network;

[0084] In this embodiment, based on the behavioral topological feature spectrum, the user basic information block is associated with the behavioral features, and the node types of the multi-layer heterogeneous network are defined, including user nodes, behavior nodes, and topological feature nodes. The information in the behavioral topological feature spectrum is used to establish the association edge between the user and the behavior. For example, by setting the weight value to reflect the intensity of the user's participation in a certain behavior, a multi-layer heterogeneous network is constructed, in which each layer represents a relationship type (such as user behavior association, behavioral topological feature association, etc.), forming a user-behavior association network, which integrates multi-dimensional association data of user information, behavioral characteristics, and topological characteristics.

[0085] Step S33: performing adjacency matrix analysis on the user-behavior association network to generate a consensus information matrix;

[0086] In this embodiment, the user-behavior association network is converted into an adjacency matrix form, where each element in the matrix represents the connection relationship and its strength between nodes in the network. The connection weights in the adjacency matrix are calculated using a graph analysis method, and the matrix is ​​normalized to eliminate the influence of data of different dimensions. The consensus information of key nodes in the network is extracted by calculating the eigenvalue decomposition or spectral clustering of the matrix. The generated consensus information matrix shows the overall association strength between users and behaviors, and reflects the collective characteristics and trends of user behaviors.

[0087] Step S34: Perform hierarchical fusion processing on the consensus information matrix to obtain user behavior consensus data.

[0088] In this embodiment, the consensus information matrix is ​​processed in layers, and the layers are divided according to different parts of the matrix (such as the basic association between users and behaviors, deep behavioral patterns, etc.). The consensus information of each layer is fused using a weighted fusion method, and information at different levels is combined to construct a comprehensive data view. For example, through weighted averaging or a multi-level fusion model, the hierarchical information is aggregated into a unified user behavior consensus data set. The generated user behavior consensus data covers the global and local characteristics of user behavior.

[0089] Preferably, step S4 comprises:

[0090] Step S41: construct a unique user identifier for the user behavior consensus data to obtain a user identifier fingerprint;

[0091] Step S42: performing corresponding unique behavior identification on the user behavior consensus data based on the user identification fingerprint to generate corresponding unique behavior data;

[0092] Step S43: performing joint feature fusion on the user identification fingerprint and the corresponding unique behavior data to obtain dynamic encryption parameters;

[0093] Step S44: Dynamically jointly encrypt the multimodal financial information based on the dynamic encryption parameters to generate encrypted financial information.

[0094] The present invention can uniquely identify each user by constructing a user unique identification for user behavior consensus data and generating a user identification fingerprint, thereby improving the authentication and verification capabilities of user identities and enhancing information security. Based on the user identification fingerprint, the user behavior consensus data is identified with corresponding unique behaviors and corresponding unique behavior data is generated, which can accurately distinguish the behavior patterns of different users and help identify the specific behavior characteristics and abnormal activities of users. By performing joint feature fusion on the user identification fingerprint and the corresponding unique behavior data, dynamic encryption parameters are obtained, which combine the user's behavior characteristics and identity information, and enhance the security and adaptability of the encryption process. Based on the dynamic encryption parameters, multimodal financial information is dynamically jointly encrypted to generate encrypted financial information, which can ensure that the financial information maintains a high level of security during storage and transmission, and prevent unauthorized access and data leakage. By generating a user unique identification fingerprint and dynamic encryption parameters, the security of the data can be significantly improved, and the data can be prevented from being maliciously tampered with or illegally accessed. The improved data security is crucial for protecting sensitive financial information. The use of dynamic encryption parameters makes the encryption process more complex and flexible, thereby preventing forgery and fraud, making it difficult to crack encrypted information, and enhancing the protection of financial transactions. By combining user identification fingerprints and behavioral data, it is possible to provide personalized security protection for each user, allowing the system to more accurately identify and respond to user-specific security needs. The joint feature fusion and dynamic encryption process optimizes the processing and protection of financial information, improves the efficiency and effectiveness of data encryption, and reduces manual intervention and processing errors through automated and intelligent processing.

[0095] In this embodiment, the user behavior consensus data is processed using a cryptographic hash algorithm (such as SHA-256). The specific operation includes extracting all the user's activity records from the user behavior consensus data, such as login time, transaction history, behavior pattern, etc., and splicing this information to form a data string. By applying the hash function, the data string is encrypted to generate a hash value of a fixed length. The hash value is the user's unique identification fingerprint, which can uniquely identify each user and ensure data consistency and privacy protection. Based on the user identification fingerprint, the fingerprint is used as input to use a machine learning classification model (such as a random forest or a support vector machine) for behavior recognition. The model training process uses the labeled user behavior data for training to generate a classifier of the user behavior pattern. After the user identification fingerprint is input, the classifier will identify the user's behavior pattern and match the identification result with the user's behavior record to generate the corresponding unique behavior data. This data specifically reflects the behavioral characteristics of each user, such as frequently operated functions and transaction habits. The user identification fingerprint and the corresponding unique behavior data are extracted and converted into numerical feature vectors. During the feature extraction process, feature engineering techniques are used, such as standardization and normalization. The two feature vectors are merged by a feature fusion algorithm (such as principal component analysis PCA or feature concatenation) to generate a joint feature vector. The vector is further processed, such as weighted processing or the final dynamic encryption parameter is generated by an encryption algorithm (such as AES). The multimodal financial information is encrypted by a symmetric encryption algorithm (such as AES or RSA) using the dynamic encryption parameter as a key. The specific operation includes dividing the multimodal financial information (such as transaction records, user personal information, etc.) into suitable blocks and applying the encryption algorithm to each block. During the encryption process, the dynamic encryption parameter is used to generate the encryption key for each data block. After the data block is encrypted, it forms encrypted financial information. This information can effectively protect data privacy during storage and transmission, ensuring that only authorized users with decryption keys can access the original financial information.

[0096] Preferably, the joint feature fusion of the user identification fingerprint and the corresponding unique behavior data includes:

[0097] Perform feature layering processing on user identification fingerprints and corresponding unique behavior data to obtain a multi-level feature map;

[0098] According to the multi-level feature map, the user identification fingerprint and the corresponding unique behavior data are mapped to obtain hierarchical correlation features;

[0099] Based on the hierarchical correlation features, the user identification fingerprint and the corresponding unique behavior data are jointly fused to obtain dynamic encryption parameters.

[0100] The present invention obtains a multi-level feature map by performing feature hierarchical processing on the user identification fingerprint and the corresponding unique behavior data, which can comprehensively extract and represent different levels of data information, thereby providing a deeper understanding of user behavior and identity. Based on the multi-level feature map, the associated feature mapping is performed to generate hierarchical associated features, which can reveal the complex relationship between the user identification fingerprint and the behavior data, and improve the accurate identification and analysis capabilities of user features. Based on the hierarchical associated features, joint feature fusion is performed to generate dynamic encryption parameters, which can comprehensively consider the multiple features of the user, generate encryption parameters with high complexity and adaptability, and improve the security of encryption. The dynamic encryption parameters generated by the joint feature fusion have high complexity and variability, enhance the encryption strength, prevent simple cracking and attacks, make the encryption process more secure through dynamic characteristics, and protect financial data from unauthorized access. The use of multi-level feature maps and hierarchical associated features improves the accuracy of data protection. By integrating the user identification and behavior data features, the generated dynamic encryption parameters can provide customized protection for the behavior patterns of different users. The generation of dynamic encryption parameters enables security policies to be adjusted according to changes in user behavior and identity characteristics, and can apply adaptive security policies in different situations to enhance the protection capabilities of financial information. The dynamic encryption parameters generated by feature fusion are unique and difficult to predict, significantly reducing the risk of forgery and fraud, and ensuring the uniqueness and security of each encryption operation. The feature fusion method combining user identification fingerprints and behavioral data enables the financial information security protection system to intelligently respond to various potential threats and improve the efficiency and effectiveness of overall security protection.

[0101] In this embodiment, the user identification fingerprint and the corresponding unique behavior data are converted into numerical feature vectors, and preliminary data preprocessing is performed, including denoising and standardization. Each feature vector is hierarchically processed, and a feature hierarchical algorithm such as an autoencoder is used to decompose the feature vector into multiple levels, including the original feature layer, the secondary feature layer, and the high-level feature layer. An autoencoder network is constructed, and the user identification fingerprint and behavior data features are input. Features are gradually extracted through multiple hidden layers to form a multi-level feature map, which shows feature representations at different levels, reflecting different abstract levels and correlation levels of the data. According to the multi-level feature map, the user identification fingerprint and the corresponding unique behavior data are associated with feature mapping to obtain hierarchical association features. Using the features of each layer in the multi-level feature map, these features are associated and mapped through feature mapping algorithms (such as similarity calculation or convolution operation), specifically including constructing an association mapping matrix, in which each element represents the similarity or correlation between two features, using similarity calculation (such as cosine similarity) to match features at different levels, mapping the features in the user identification fingerprint and behavior data according to the degree of association, and the generated hierarchical association feature matrix reflects the association relationship between different features in the hierarchy and shows the complex interaction between data features. The hierarchical association feature matrix is ​​processed and used as the input of joint feature fusion. The features of different levels are combined using a fusion algorithm (such as weighted summation or fusion network) to construct a fusion network. The network adopts a weighted average mechanism to fuse the hierarchical association features according to the preset weights. The weights are dynamically adjusted according to the importance of the features. The network combines the features of the user identification fingerprint and the features of the behavior data to generate a unified feature vector as a dynamic encryption parameter.

[0102] Preferably, step S5 comprises:

[0103] Step S51: constructing an attack strategy network for the predicted behavior data to generate a predicted attack strategy network;

[0104] Step S52: Evolving adversarial samples on encrypted financial information according to the predicted attack strategy network to obtain simulated attack data;

[0105] Step S53: simulating a vulnerability attack on the encrypted financial information based on the simulated attack data to generate simulated attack financial information;

[0106] Step S54: Perform security threat screening on the simulated attack financial information to obtain security threat data.

[0107] The present invention constructs an attack strategy network for predicted behavior data to generate a predicted attack strategy network, which can simulate and predict potential attack strategies, provide a comprehensive analysis of various attack methods faced by financial information systems, and improve the ability to identify attack patterns. According to the attack strategy network, encrypted financial information is subjected to adversarial sample evolution to obtain simulated attack data, which can help simulate attack scenarios, thereby testing and strengthening the anti-attack ability of encrypted financial information. Based on the simulated attack data, vulnerability attack simulation is performed on encrypted financial information to generate simulated attack financial information, which can identify potential vulnerabilities in encrypted financial information, help evaluate the security protection effect of the system, and reveal the security defects of the system. Security threat screening is performed on simulated attack financial information to obtain security threat data, which can accurately detect and evaluate security threats in financial information systems, and provide targeted security protection suggestions and improvement measures. By constructing an attack strategy network and simulating attacks, potential security threats can be identified and predicted in advance, which helps to take protective measures before real attacks occur, thereby improving the security of the system. Evolution of adversarial samples and simulation of vulnerability attacks on encrypted financial information can strengthen the defense capability of the encryption system. Through simulated attacks and tests, it is ensured that encryption technology can resist various attacks and improve the security and reliability of the system. By analyzing security threat data, we can optimize existing security protection strategies and develop more effective countermeasures, which will help better protect financial information in actual attacks and reduce potential security risks. Vulnerability attack simulation can help discover and repair security vulnerabilities in the system, reduce the security risks caused by vulnerabilities, and ensure the security of the system before it is put into use. By comprehensively analyzing attack strategies and simulating attack results, we can improve the ability and level of risk management, help develop comprehensive security strategies, and ensure the security and integrity of financial information.

[0108] In this embodiment, feature extraction is performed on the collected prediction behavior data. The important features in the data are extracted through feature engineering methods, such as principal component analysis (PCA) or feature selection algorithms. A graph structure learning algorithm, such as a network generation method based on game theory or a generative adversarial network (GAN), is used to construct an attack strategy network. In the network construction process, each node represents a potential attack strategy, and the connection between nodes represents the association or dependency relationship between different strategies. The network is adjusted and optimized through an iterative optimization algorithm, so that the generated attack strategy network can effectively simulate the potential attack paths and strategy combinations in the real environment, and finally a predicted attack strategy network is obtained. Based on the attack strategy network, evolutionary algorithms or genetic algorithms are used to generate adversarial samples for encrypted financial information. The specific operation is to initialize a group of adversarial samples. These samples are generated by perturbing specific features in the original encrypted financial information. Driven by the evolutionary algorithm, the sample group undergoes multiple iterations. In each iteration, the fitness of the samples is evaluated and selected based on the attack strategy network, and samples with stronger attack effects are retained. A new generation of adversarial samples is generated through crossover and mutation operations. This process continues until the evolved adversarial samples can successfully attack the security mechanism of the original encrypted financial information. The generated simulated attack data is used to simulate the attack behavior that occurs in the real attack scenario. Based on the obtained simulated attack data, a penetration testing tool or a custom vulnerability scanner is used to simulate vulnerability attacks on encrypted financial information. The simulated attack data is input into the vulnerability scanner. The scanner conducts in-depth analysis of the encrypted financial information based on the preset vulnerability rule library and simulated attack strategy to identify potential security vulnerabilities. In the analysis process, the encryption algorithm, key management mechanism, and data transmission path are comprehensively checked by combining static analysis technology and dynamic analysis technology. After the vulnerability attack simulation results are analyzed, the generated simulated attack financial information describes in detail the vulnerabilities and security risks exposed by the encrypted financial information in the simulated attack scenario. The simulated attack financial information is classified and labeled using a pre-trained machine learning model to identify different types of security threats. A rule-based security audit system is applied to further verify and screen the identified threats. The audit system detects whether there are known or unknown threat patterns in the simulated attack financial information by matching the existing threat library. For unknown threats, anomaly detection-based algorithms, such as isolation forests or autoencoders, are used to perform behavioral anomaly analysis. By combining the threat labels output by the machine learning model and the analysis results of the security audit system, the most threatening security risks are screened out to form security threat data to ensure a comprehensive grasp of potential security issues.

[0109] Preferably, the security threat screening of simulated attack financial information includes:

[0110] Perform time series conversion on the simulated attack financial information to obtain simulated time series data;

[0111] Conduct change trend analysis on simulated time series data to generate protection change trends;

[0112] Based on the protection change trend, the simulation time series data is changed to locate the data change point and obtain the safety data change point;

[0113] Security threat screening is performed on the simulated attack financial information based on the security data change points to obtain security threat data.

[0114] The present invention can accurately capture the information change process from the time dimension by performing time series conversion on the simulated attack financial information to obtain simulated time series data, laying the foundation for subsequent trend analysis. The simulated time series data is analyzed for change trends to generate protection change trends, which can identify the change trends of security protection during the attack process, deeply understand how the attack behavior affects the security of financial information, and help to adjust the protection strategy in time. Based on the protection change trend, the simulated time series data is located for change data to obtain the security data change point, which can accurately locate the key change point of information security, help to quickly identify potential security threat areas, and reduce the time and resource consumption of analysis. According to the security data change point, the simulated attack financial information is screened for security threats to obtain security threat data, which can accurately identify security loopholes or weaknesses in the attack behavior, provide specific data support for the improvement of the defense system, and improve the accuracy of screening. Through time series conversion and trend analysis, real-time security monitoring of the financial information system can be achieved, which helps to effectively intervene when the security threat just appears to prevent the expansion of the threat and potential losses. Change trend analysis can provide forward-looking threat prediction capabilities. By analyzing the changes in trends, future security incidents can be foreseen and preventive measures can be taken, which helps to improve the overall security of the system. By accurately locating the changing data, security protection measures can be adjusted and strengthened according to the specific security data change points, achieving refined security protection, reducing unnecessary resource waste, and improving protection efficiency. Through the analysis of time series data and the identification of change points, the security protection system can dynamically respond to changing security threats, ensuring that the system can adapt to changes in different attack scenarios and maintain efficient protection capabilities. Accurate location of changing data and threat screening reduce false alarm rates, avoid unnecessary security alerts, and improve the effectiveness and credibility of the security protection system.

[0115] In this embodiment, all data points in the simulated attack financial information are obtained, and these data points are rearranged in chronological order through the time series analysis method. Time series data mining technology, such as sliding window algorithm, is used to convert the characteristic data in the simulated attack financial information into continuous time series data to form an ordered time series, and the time dimension is extracted and converted to ensure that each data point reflects its evolution trend on the time axis. After the time series conversion is completed, the obtained simulated time series data will be used for subsequent change trend analysis. For the generated simulated time series data, trend analysis tools are used for in-depth analysis. By applying time series analysis algorithms such as autoregressive integrated moving average model (ARIMA) or Holt-Winters exponential smoothing method, the change trend in the data is analyzed to capture the growth, decline, fluctuation and other characteristics of the data in different time periods. During the analysis process, the significance of the trend is statistically tested to ensure the reliability of the analysis results. The analysis results are summarized to generate a complete protection change trend line. According to the generated protection change trend, change point detection algorithms such as cumulative sum control chart (CUSUM) or Bayesian change point detection method are used to gradually scan the simulated time series data to locate the time points when the data changes significantly. Each change point indicates the start time of the existing security threat or abnormal activity. Combined with the slope change of the trend line and the amplitude of data fluctuation, multi-level positioning and verification are carried out to ensure that each identified change point has practical security significance. After positioning is completed, all change points are summarized as basic data for subsequent threat screening. Comprehensive security threat screening of simulated attack financial information is carried out using security data change points. The screening process uses rule-based automated threat detection tools combined with machine learning models for fine-grained analysis. Specific operations include comparing and matching the data features of each change point with known threat patterns, evaluating its threat level, and using anomaly detection algorithms to conduct in-depth analysis of data near the change point to identify potential unknown threats.

[0116] Preferably, step S6 comprises:

[0117] Step S61: Risk stratify the encrypted financial information according to the security threat data to obtain multi-level risk sub-blocks;

[0118] Step S62: performing cyclic homomorphic encryption on the multi-level risk sub-blocks to obtain homomorphically encrypted sub-blocks;

[0119] Step S63: Perform random nested encryption on the security threat data based on the homomorphic encryption sub-block to generate multi-layer encrypted financial information;

[0120] Step S64: logically encapsulate the multi-layer encrypted financial information to generate re-encrypted financial information.

[0121] The present invention performs risk stratification on encrypted financial information according to security threat data to obtain multi-level risk sub-blocks, which can finely divide financial information according to risk levels, help to take targeted protective measures for risks of different levels, and improve the accuracy of risk management. The multi-level risk sub-blocks are cyclically homomorphically encrypted to obtain homomorphically encrypted sub-blocks. The cyclic homomorphic encryption technology allows data to be operated without decryption, which not only ensures the security of the data, but also supports the secure calculation and analysis of the data. Based on the homomorphically encrypted sub-blocks, the security threat data is randomly nested and encrypted to generate multi-layer encrypted financial information, which further improves the complexity and security of the encrypted data, increases the difficulty of cracking by attackers, and effectively prevents data leakage. Through the multi-level encryption method, the protection level of the data is increased, so that even if some encryption layers are broken, there are still other encryption layers to provide protection, effectively resisting complex attacks and improving the overall data security. The multi-layer encrypted financial information is logically encapsulated to generate re-encrypted financial information, which ensures the integrity and non-tamperability of the encrypted data during transmission and storage, and provides higher security for the data. By performing layered encryption and encapsulation of data at different risk levels, it can dynamically adapt to different security needs and risk scenarios, improving the adaptability and flexibility of financial information security protection. By implementing multi-level encryption and logical encapsulation strategies, the overall security of the system has been significantly improved, and it can effectively respond to various potential security threats.

[0122] In this embodiment, by analyzing security threat data, risk assessment is performed on different data parts in the encrypted financial information, and a weighted scoring method or risk matrix is ​​used to quantify the risk level of each data part, generating a hierarchical structure containing high, medium and low risk levels. Using a recursive partitioning algorithm, the encrypted financial information is segmented according to the assessment results to form multiple risk sub-blocks with independent risk levels. The content of each risk sub-block is strictly divided according to its corresponding threat level and impact range, ensuring that different levels of protection and encryption can be carried out in a targeted manner in the subsequent processing process. Each sub-block is encrypted using a cyclic homomorphic encryption algorithm. The specific steps include generating a corresponding public key and private key pair for each risk sub-block, and dynamically adjusting the parameters of the homomorphic encryption according to the risk level, such as the degree of the encryption polynomial or the size of the modulus. These encryption parameters are applied one by one to cyclically encrypt the data of the risk sub-blocks to ensure that the encrypted data can be operated while achieving security isolation of different risk levels. After each sub-block is encrypted, a corresponding homomorphic encryption sub-block is generated, and these encrypted sub-blocks are used to perform random nested encryption on the security threat data. During the operation, a set of nested encryption sequences is generated by a random number generator, and the security threat data is encrypted layer by layer based on this sequence. The security threat data is divided into several parts, and each part is nested in the corresponding homomorphic encryption sub-block for initial encryption. The nested encrypted data is further encrypted at multiple levels to ensure that in the encryption chain, each layer of encryption uses different homomorphic encryption sub-blocks and parameters to form a complex and difficult-to-crack multi-layer encryption structure. The resulting multi-layer encrypted financial information reaches a high level of protection in terms of security and can effectively resist various types of attacks. The structure of encrypted information is optimized by using orchestration logic algorithms, redundant information is reduced, and the distribution of encryption levels is optimized. The logical structure of encrypted information is integrated using data encapsulation technology to form a closed encryption system. By setting access control policies and calling rules for encrypted information, the protection capabilities of information are further enhanced, and heavily encrypted financial information is generated. It is ensured that the data can still be accessed and operated through specific decryption paths in the logically encapsulated state, thereby achieving the goal of both ensuring security and maintaining data availability.

[0123] The present invention also provides an artificial intelligence-based financial information security protection system, which is used to execute the artificial intelligence-based financial information security protection method as described above. The artificial intelligence-based financial information security protection system includes:

[0124] The information acquisition module is used to acquire multimodal financial information; identify the modal type of the multimodal financial information, and group the information blocks to obtain the financial transaction behavior information block and the user basic information block;

[0125] The behavior prediction module is used to analyze the behavior patterns of the financial transaction behavior information blocks, predict the behavior trends, and generate predicted behavior data;

[0126] The information mapping module is used to perform hierarchical association mapping on the user basic information blocks based on the predicted behavior data to obtain the user behavior consensus data;

[0127] Joint encryption module, used to dynamically joint encrypt multimodal financial information using user behavior consensus data to generate encrypted financial information;

[0128] A threat screening module is used to simulate vulnerability attacks on encrypted financial information based on predicted behavior data, and to perform security threat screening to obtain security threat data;

[0129] The secondary encryption module is used to perform secondary encryption on encrypted financial information based on security threat data to generate re-encrypted financial information.

[0130] The present invention can identify and classify different types of multimodal financial information by utilizing the information acquisition module, divide complex financial data into behavioral information blocks and basic information blocks, and enhance the accuracy and efficiency of the system in processing multidimensional data. The behavioral prediction module analyzes the financial transaction behavior pattern, performs trend prediction, and identifies potential risks and abnormal behaviors in advance, thereby improving the system's foresight and responsiveness. The information mapping module uses the predicted behavior data to perform hierarchical association mapping and generates user behavior consensus data, so that the association between user basic information and behavior information is closer and more accurate, providing a solid foundation for subsequent encryption and protection. The joint encryption module uses dynamic joint encryption technology to adjust the encryption strategy in real time according to the user behavior consensus data, making the encryption process more flexible and able to effectively respond to changing security threats. The threat screening module simulates vulnerability attacks and performs security threat screening to timely discover potential threats in encrypted information, which helps to identify and prevent security vulnerabilities in advance and enhance the overall protection capability of the system. The secondary encryption module performs secondary encryption in combination with security threat data on the basis of the initial encryption, further improving the confidentiality and security of financial information, and ensuring that high-intensity protection can be maintained in the face of complex attacks.

[0131] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0132] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A financial information security protection method based on artificial intelligence, characterized in that: The following steps are involved: Step S1: Obtain multimodal financial information; Identify the modal type of multimodal financial information and group the information blocks to obtain financial transaction behavior information blocks and user basic information blocks; Step S2: Analyze the behavior pattern of the financial transaction behavior information block, predict the behavior trend, and generate predicted behavior data; wherein step S2 includes the following steps: Step S21: Decomposing the financial transaction behavior information block into time series features to obtain a time series feature sequence; Step S22: performing subsequence pattern matching on the financial transaction behavior information block according to the time series feature sequence to generate a behavior pattern matching set; Step S23: performing center point clustering processing on the behavior pattern matching set to obtain pattern center data; Step S24: Analyze the change trend of the pattern center data to generate pattern trend data; Step S25: predicting the behavior trend of the financial transaction behavior information block based on the pattern trend data to generate predicted behavior data; Step S3: Based on the predicted behavior data, hierarchical association mapping is performed on the user basic information blocks to obtain user behavior consensus data; wherein step S3 includes the following steps: Step S31: performing continuous coherence analysis on the predicted behavior data to obtain a behavior topology feature spectrum; Step S32: construct a multi-layer heterogeneous network for the user basic information block according to the behavior topology feature spectrum to obtain a user-behavior association network; Step S33: performing adjacency matrix analysis on the user-behavior association network to generate a consensus information matrix; Step S34: performing hierarchical fusion processing on the consensus information matrix to obtain user behavior consensus data; Step S4: dynamically jointly encrypt multimodal financial information using user behavior consensus data to generate encrypted financial information; wherein step S4 includes the following steps: Step S41: construct a unique user identifier for the user behavior consensus data to obtain a user identifier fingerprint; Step S42: performing corresponding unique behavior identification on the user behavior consensus data based on the user identification fingerprint to generate corresponding unique behavior data; Step S43: performing joint feature fusion on the user identification fingerprint and the corresponding unique behavior data to obtain dynamic encryption parameters; wherein step S43 specifically includes the following steps: Perform feature layering processing on user identification fingerprints and corresponding unique behavior data to obtain a multi-level feature map; According to the multi-level feature map, the user identification fingerprint and the corresponding unique behavior data are mapped to obtain hierarchical correlation features; Based on the hierarchical correlation features, the user identification fingerprint and the corresponding unique behavior data are combined to obtain dynamic encryption parameters; Step S44: dynamically jointly encrypting the multimodal financial information based on the dynamic encryption parameters to generate encrypted financial information; Step S5: simulating vulnerability attacks on encrypted financial information based on the predicted behavior data, and performing security threat screening to obtain security threat data; Step S6: re-encrypt the encrypted financial information based on the security threat data to generate re-encrypted financial information.

2. The financial information security protection method based on artificial intelligence according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: acquiring multimodal financial information; performing feature extraction on the multimodal financial information to obtain a multidimensional feature vector; Step S12: performing modal clustering analysis on the multimodal financial information according to the multidimensional feature vector to generate a modal clustering result; Step S13: performing feature recognition on the modal clustering results class by class to obtain a class modal feature vector; Step S14: performing modal type identification on the category modal feature vector to generate modal classification data; Step S15: Group the multimodal financial information into information blocks based on the modal classification data to obtain financial transaction behavior information blocks and user basic information blocks.

3. The financial information security protection method based on artificial intelligence according to claim 1 is characterized in that: Step S5 includes the following steps: Step S51: constructing an attack strategy network for the predicted behavior data to generate a predicted attack strategy network; Step S52: Evolving adversarial samples on encrypted financial information according to the predicted attack strategy network to obtain simulated attack data; Step S53: simulating a vulnerability attack on the encrypted financial information based on the simulated attack data to generate simulated attack financial information; Step S54: Perform security threat screening on the simulated attack financial information to obtain security threat data.

4. The financial information security protection method based on artificial intelligence according to claim 3 is characterized in that: Step S54 specifically includes the following steps: Perform time series conversion on the simulated attack financial information to obtain simulated time series data; Conduct change trend analysis on simulated time series data to generate protection change trends; Based on the protection change trend, the simulation time series data is changed to locate the data change point and obtain the safety data change point; Security threat screening is performed on the simulated attack financial information according to the security data change points to obtain security threat data.

5. The financial information security protection method based on artificial intelligence according to claim 1 is characterized in that: Step S6 includes the following steps: Step S61: Risk stratify the encrypted financial information according to the security threat data to obtain multi-level risk sub-blocks; Step S62: performing cyclic homomorphic encryption on the multi-level risk sub-blocks to obtain homomorphically encrypted sub-blocks; Step S63: Perform random nested encryption on the security threat data based on the homomorphic encryption sub-block to generate multi-layer encrypted financial information; Step S64: logically encapsulate the multi-layer encrypted financial information to generate re-encrypted financial information.

6. A financial information security protection system based on artificial intelligence, characterized in that: Used to execute the financial information security protection method based on artificial intelligence as claimed in claim 1, the financial information security protection system based on artificial intelligence includes: The information acquisition module is used to acquire multimodal financial information; identify the modal type of the multimodal financial information, and group the information blocks to obtain the financial transaction behavior information block and the user basic information block; The behavior prediction module is used to analyze the behavior patterns of the financial transaction behavior information blocks, predict the behavior trends, and generate predicted behavior data; The information mapping module is used to perform hierarchical association mapping on the user basic information blocks based on the predicted behavior data to obtain the user behavior consensus data; Joint encryption module, used to dynamically joint encrypt multimodal financial information using user behavior consensus data to generate encrypted financial information; A threat screening module is used to simulate vulnerability attacks on encrypted financial information based on predicted behavior data, and to perform security threat screening to obtain security threat data; The secondary encryption module is used to perform secondary encryption on encrypted financial information based on security threat data to generate re-encrypted financial information.

Images