Intelligent bank payment and settlement method supporting real-time risk control and multi-channel interaction

Abstract

The invention relates to an intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction. The method comprises the following steps: performing multi-dimensional association analysis on multi-channel transaction behavior flow data to obtain transaction channel associated data, and performing deep feature learning on the transaction channel associated data to obtain a transaction behavior feature vector; performing protocol conversion and field mapping processing on the multi-channel original message data to obtain standard payment transaction data; performing random network modeling on the multi-channel transaction processing flow according to the standard payment transaction data to obtain a processing time delay threshold parameter; and constructing a state space and a reward function of the TR-A2C deep reinforcement learning network according to the transaction behavior feature vector and the processing time delay threshold parameter, and obtaining a multi-channel real-time risk control decision strategy through parallel strategy optimization. According to the invention, distributed storage and dynamic scheduling of multi-channel data are realized, the problems of system expansibility and data access performance are solved, and the accuracy and real-time performance of risk control decision are improved.

Description

Technical Field

[0001] The present invention relates to the field of payment and settlement technology, and in particular to an intelligent bank payment and settlement method supporting real-time risk control and multi-channel interaction. Background Art

[0002] Bank payment and settlement systems are facing challenges such as a surge in multi-channel transaction data, increased real-time risk control requirements, and system processing performance bottlenecks. Traditional payment and settlement systems generally have problems such as inconsistent channel data standards, rigid risk control rules, and system response delays, making it difficult to meet business needs.

[0003] In multi-channel payment scenarios, transaction data formats and communication protocols of different channels vary greatly, making data integration difficult and risk control rules difficult to manage and dynamically adjust. At the same time, as the scale of transactions grows, the system needs to complete risk control decisions at the millisecond level, which places higher demands on computing performance and processing latency. Summary of the invention

[0004] The main purpose of the present invention is to provide an intelligent bank payment settlement method that supports real-time risk control and multi-channel interaction. The present invention realizes distributed storage and dynamic scheduling of multi-channel data, solves the problems of system scalability and data access performance, and improves the accuracy and real-time performance of risk control decisions.

[0005] To achieve the above object, the present invention provides an intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction, comprising the following steps: Performing multi-dimensional correlation analysis on multi-channel transaction behavior flow data to obtain transaction channel correlation data, and performing deep feature learning on the transaction channel correlation data to obtain a transaction behavior feature vector; Perform protocol conversion and field mapping on original message data from multiple channels to obtain standard payment transaction data; Performing random network modeling on the multi-channel transaction processing flow according to the standard payment transaction data to obtain a processing delay threshold parameter; The state space and reward function of the TR-A2C deep reinforcement learning network are constructed according to the transaction behavior feature vector and the processing delay threshold parameter, and a multi-channel real-time risk control decision strategy is obtained through parallel strategy optimization.

[0006] The present invention also provides an intelligent bank payment and settlement device supporting real-time risk control and multi-channel interaction, comprising: A correlation analysis unit, used to perform multi-dimensional correlation analysis on multi-channel transaction behavior flow data to obtain transaction channel correlation data, and perform deep feature learning on the transaction channel correlation data to obtain a transaction behavior feature vector; A conversion unit, used to perform protocol conversion and field mapping processing on multi-channel original message data to obtain standard payment transaction data; A modeling and analysis unit, configured to perform random network modeling on a multi-channel transaction processing flow according to the standard payment transaction data to obtain a processing delay threshold parameter; The risk control decision unit is used to construct the state space and reward function of the TR-A2C deep reinforcement learning network according to the transaction behavior feature vector and the processing delay threshold parameter, and obtain a multi-channel real-time risk control decision strategy through parallel strategy optimization.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention uses multi-dimensional correlation analysis and deep feature learning to uniformly model and extract features of multi-channel transaction behavior data, solves the problem of inconsistent data standards in different channels, and improves the ability to express features. The multi-layer Transformer attention network and TR-A2C deep reinforcement learning algorithm are used to achieve dynamic optimization and adaptive adjustment of risk control strategies, breaking through the limitations of traditional static rule engines. Based on the random network calculation method, the transaction processing delay is accurately modeled, combined with a double-layer cache mechanism, to ensure the real-time response capability of the system in high-concurrency scenarios. Through the consistent hashing algorithm and load balancing strategy, the distributed storage and dynamic scheduling of multi-channel data are realized, solving the problems of system scalability and data access performance. A complete data standardization processing flow is designed, including protocol conversion, field mapping and data verification, to ensure the consistency and reliability of multi-channel data. Through parallel strategy optimization and multi-environment sampling mechanism, the accuracy and real-time performance of risk control decisions are improved, and the risk of transaction fraud is effectively prevented. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of the steps of an intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction in one embodiment of the present invention; Figure 2 It is a structural block diagram of an intelligent bank payment and settlement device supporting real-time risk control and multi-channel interaction in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] 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

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 This embodiment provides an intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction, including the following steps: S1, conduct multi-dimensional correlation analysis on multi-channel transaction behavior flow data to obtain transaction channel correlation data, and conduct deep feature learning on the transaction channel correlation data to obtain transaction behavior feature vectors; Among them, the different dimensions of multi-channel transaction behavior flow data are specifically processed in order to effectively extract valuable feature information. In the time dimension of transaction data, the time slot division method is adopted, that is, the transaction data is sliced ​​according to the preset time interval (such as minute level, hour level or day level), and the transaction is mapped to the corresponding time slot according to the time point of the transaction, forming a binary feature matrix, in which each time slot corresponds to a binary value, indicating whether a transaction occurred in the time period, thereby effectively capturing the distribution law of transactions in time. At the same time, in order to process the numerical information in the transaction data, such as transaction amount, number of transactions or transaction frequency, it is discretized by using the unique hot encoding method, that is, according to the preset numerical interval, the continuous numerical data is divided into multiple intervals, and an independent binary feature is created for each interval to form a numerical feature distribution matrix, thereby avoiding learning bias caused by the uneven distribution of numerical data, and enhancing the model's perception of information such as transaction amount. When processing the geographic location information in the transaction data, the spatial distribution characteristics of the transaction behavior are measured by calculating the Euclidean distance between the places where the transaction occurs. In order to eliminate the influence of different geographical scales, the calculated Euclidean distance is normalized to obtain a location similarity matrix, which reflects the proximity between different transactions in geographical space, thereby helping to discover abnormal cross-regional transaction behaviors or potential fraud patterns. In order to improve the risk control ability, the device identification information is feature encoded, and the device fingerprint technology is used to extract stable features from the device's IP address, browser fingerprint, operating system version, screen resolution and other information, and hash code or embed mapping is performed to obtain the device fingerprint feature matrix, which effectively characterizes the uniqueness of the device. The binary feature matrix, numerical feature distribution matrix, location similarity matrix and device fingerprint feature matrix are subjected to feature contribution calculation and data association analysis to extract the key features of multi-channel transaction behavior. The feature contribution calculation adopts a method based on information gain or Shapley value to measure the influence of various features on the prediction of transaction behavior, thereby screening out the most distinguishing feature dimensions, while the data association analysis adopts methods such as association rule mining or graph neural network to mine the intrinsic relationship between different features and obtain transaction channel association data. The transaction channel association data is input into the multi-layer Transformer attention network for deep feature learning. The Transformer network dynamically weights the input data through the self-attention mechanism to capture the long-distance dependency between transaction behavior features, while avoiding the defects of traditional sequence models in modeling long-term dependencies. The multi-layer Transformer structure stacks multiple self-attention layers, and each layer performs multi-head attention calculations on the input data, so that the information between different transaction features interacts on different attention heads, and performs nonlinear transformations through feedforward neural networks to enhance the expressiveness of features.After learning through multiple layers of Transformer networks, a transaction behavior feature vector is obtained, which is used for training and reasoning of risk control models to support real-time risk control decisions.

[0014] The transaction channel association data is input into the multi-head attention layer, and features are extracted through the multi-head attention mechanism. That is, in the multi-head attention layer, 8 parallel attention heads are used to calculate the dot product of the query matrix, key matrix and value matrix to obtain the attention weight distribution. Each attention head models the input data from different angles to capture the dependencies of transaction behaviors at different levels. The attention weight distribution is scaled to prevent the value from being too large or too small to affect the learning stability. The dot product result is adjusted by the scaling factor, and then the softmax normalization is applied to ensure that the sum of all attention weights is 1 to obtain a probability distribution, so that the model can more reasonably allocate the weights of different transaction features. The normalized attention weight distribution is weighted and summed with the value matrix to obtain the multi-head attention feature. The multi-head attention feature is input into a two-layer feedforward neural network to learn the deep features of the transaction behavior. In the first layer of the feedforward neural network, the ReLU activation function is used to perform nonlinear transformation on the input data, so that the model can learn the complex patterns of transaction behaviors and enhance the nonlinear expression ability of the features, thereby improving the risk control system's ability to identify abnormal transactions. In the second layer of feedforward neural network, linear transformation is used to map the intermediate features to obtain output features, which contain the deep representation of trading behavior. In order to avoid the problem of gradient vanishing or gradient exploding and improve the training stability of the network, the output features are residually connected, and the input features are added element by element to the output features after the feedforward neural network transformation, so that the gradient can better retain the initial information during back propagation and ensure the training effect of the model. The features after residual connection are layer normalized to eliminate the distribution differences between different feature dimensions, making the network training more stable and improving the generalization ability of the model, and obtaining the output feature vector of the first layer Transformer block. The output feature vector of the first layer Transformer block is sequentially input into the second layer Transformer block and the third layer Transformer block. In each layer Transformer block, multi-head attention calculation, feedforward neural network transformation and layer normalization are repeatedly performed to ensure that each layer fully learns the deep features of trading behavior. Through the design of multi-layer structure, the model captures the multi-scale features of trading data at different levels and gradually learns the long-term dependencies of trading behavior. The multi-layer feature representation is subjected to feature fusion and dimensionality reduction processing to extract the most representative transaction behavior feature vector. Feature fusion uses weighted summation or attention aggregation to integrate feature information from different layers so that the final feature vector can more comprehensively represent the transaction behavior. Dimensionality reduction processing uses principal component analysis or fully connected layer mapping to reduce feature dimensions while retaining the most important feature information, and finally obtains the transaction behavior feature vector.

[0015] S2, performing protocol conversion and field mapping processing on the original message data from multiple channels to obtain standard payment transaction data; Specifically, the protocol type analysis is performed on the original message data of multiple channels, and a protocol type identification set is generated to determine the protocol category to which each message data belongs. By analyzing the protocol header, data structure and field format, the protocol type of the message is accurately identified, and it is classified and stored in the protocol type identification set. According to the protocol type identification set, the original message data of multiple channels in different protocol formats are parsed, structured data fields are split out, and an original field data set is generated, which contains various transaction-related information, such as transaction amount, transaction time, merchant number, card number, user identity information and other original fields. The original field data set is input into the field mapping rule library for matching to determine how to convert the fields under different protocols into standardized field structures. The field mapping rule library is a set of predefined mapping relationship tables that store the mapping rules between the fields of different protocols and the standard payment data format. According to the standard field mapping table, the original field data set is converted into fields, and the field names and data formats under different protocols are uniformly converted into standard formats to ensure that all payment transaction data follow the same field structure in subsequent processing. After this conversion step, the initial standard data is obtained. Perform data integrity checks on the initial standard data to check whether there are missing fields, incorrect data formats, or empty key fields. For example, the transaction amount field must be a positive number, the transaction time field must conform to the standard time format, and the merchant number must meet specific numbering rules. Perform consistency checks to check whether the data conforms to business rules. Summarize the results of the integrity and consistency checks into the verification result data. Perform data completion and cleaning on the initial standard data based on the verification result data. If some fields are missing but can be completed by deduction, such as the transaction time is missing but recovered based on log records, these data are automatically filled in. Data cleaning is used to remove redundant or erroneous data, such as correcting incorrectly formatted amount fields, removing invalid characters, or standardizing telephone number formats, etc., to ultimately obtain a data format that meets the standards and forms standard payment transaction data.

[0016] S3, perform random network modeling on the multi-channel transaction processing flow based on standard payment transaction data to obtain processing delay threshold parameters; It should be noted that the payment transaction processing flow is divided into a data preprocessing module, a feature extraction module, a risk control decision module, and a response generation module. According to the functions and data flows of these modules, the processing node information in the standard payment transaction data is hierarchically divided to form a multi-level service node structure to construct the topological relationship of the processing nodes. By dividing these service nodes into layers, the flow path of the payment transaction data in the entire system is described to ensure that the processing logic between different modules is clear. According to the processing node topology, the data flow relationship between each node is modeled by Markov chain to construct the state transition matrix of the system. Markov chain modeling can describe the flow probability of payment transaction data between each processing node and quantify the dependency relationship between different nodes through the state transition matrix. Each processing node is regarded as a state, and the flow probability of payment transaction data between different states is determined by the state transition matrix. Based on a large amount of historical transaction data, the possibility of different processing paths is estimated to generate a node dependency graph. The relationship graph reflects the logical connection between each processing node and provides the probability weight of different transaction process paths. The historical processing data of each node in the processing node topology is estimated by maximum likelihood and the service time distribution parameters are fitted to obtain the service time probability density function of each processing node, which describes the time distribution characteristics of transaction data on different processing nodes. The node service time probability density function is Laplace transformed to calculate the cumulative delay effect of different nodes in the frequency domain space. The end-to-end delay distribution function is obtained by calculating the cumulative moment of the service time distribution, which describes the completion probability of the entire transaction process at different time points. According to the preset service quality requirements, the probability integral calculation of the end-to-end delay distribution function is performed to determine the delay constraints that need to be met for transaction processing. For example, if the system needs to ensure that 99% of the transactions can be completed within 200 milliseconds, the time threshold that meets this requirement is calculated based on the end-to-end delay distribution function. According to the delay constraints and the node dependency graph, a numerical solution is performed to calculate the optimal processing delay upper bound of the entire system. The numerical solution adopts linear programming or convex optimization methods, with the goal of minimizing the maximum processing delay while satisfying all delay constraints. In this process, by optimizing the delay calculation of different paths, the optimal delay allocation strategy for transaction processing is obtained, and finally the processing delay threshold parameters of the entire payment transaction process are calculated.

[0017] S4, constructs the state space and reward function of the TR-A2C deep reinforcement learning network based on the transaction behavior feature vector and processing delay threshold parameters, and obtains the multi-channel real-time risk control decision-making strategy through parallel strategy optimization.

[0018] Specifically, the transaction behavior feature vector is decomposed according to multiple dimensions such as transaction value, transaction frequency, transaction location and transaction device. Each dimension represents a different type of transaction feature. The transaction value reflects the trend of transaction amount, the transaction frequency describes the consumption habits of users, the transaction location captures the spatial distribution pattern of transactions, and the transaction device provides device fingerprint features to identify the stability and abnormal changes of the device. In order to consider the processing efficiency of transactions, the delay monitoring vector is constructed in combination with the processing delay threshold parameter. The monitoring vector reflects the delay of each node in the transaction process to ensure that the system effectively monitors transactions with excessive delays. The divided transaction feature subspace and the delay monitoring vector are fused by the multi-dimensional tensor splicing method to construct the state space embedding vector. The state space embedding vector is subjected to sparsity constraints and regularization to ensure that the model is not disturbed by redundant features and focuses more on key transaction features. The sparsity constraint is implemented by L1 regularization to reduce the influence of irrelevant features, while the regularization process uses batch normalization or layer normalization to keep the features within a reasonable numerical range, thereby preventing gradient vanishing or gradient explosion. The standardized state space is mapped to the risk control action space, that is, the executable operation set of the risk control strategy is defined according to the state characteristics of the transaction, such as directly approving the transaction, further identity verification, triggering risk control rules or rejecting the transaction, etc., to form an action space set, so that the reinforcement learning model can make the best decision under different transaction states. When constructing the TR-A2C deep reinforcement learning network, the standardized state space is used as input, and the network structure is designed to improve learning efficiency. The Transformer encoding layer is used to deeply represent the state space, the long-distance dependencies between transaction features are captured through the self-attention mechanism, and the interaction pattern between different features is learned using the multi-head attention mechanism to generate high-quality state representation. The output of the Transformer encoding layer is connected to the strategy layer and the value evaluation layer, where the strategy layer is used to generate the probability distribution of risk control decisions, and the value evaluation layer is used to calculate the long-term return of state-action pairs. By connecting these three modules in series, the TR-A2C deep reinforcement learning network is formed, so that risk control decision training is performed under the reinforcement learning framework. During the training process, parallel environment sampling is performed to improve data collection efficiency and accelerate strategy optimization. Create multiple trading simulation environments and run the current strategy in each environment to collect interaction data. Each environment interacts according to the transaction state space and action space, and generates execution records of transaction decisions, which are synchronously collected and stored as a strategy trajectory sequence dataset. This dataset contains information such as transaction status, execution actions, transaction processing time, and risk control feedback. The weighted combination of delay loss and risk control loss is calculated based on the strategy trajectory sequence dataset to optimize the reinforcement learning strategy. Among them, delay loss is used to measure the time cost of transaction processing, while risk control loss is used to evaluate the safety of decision-making.In order to strike a balance between risk control accuracy and transaction processing efficiency, a weighted combination method is used to calculate the overall loss function, and based on the loss, the advantage function is calculated to estimate the gain of each decision relative to the average decision, and obtain the policy gradient. The policy gradient is input into the Adam optimizer for parameter update, and the risk control decision-making ability of the model is continuously optimized through cyclic iteration until it converges to the optimal strategy, and finally a multi-channel real-time risk control decision strategy is obtained, so that the system can intelligently identify abnormal transactions and make the optimal risk control decision in real-time transaction processing.

[0019] The access frequency statistics and data volume analysis of multi-channel transaction behavior flow data and multi-channel original message data are performed to determine the access patterns and storage requirements of different types of data. By analyzing the access frequency of transaction behavior flow data, it is identified which data is frequently accessed in a short period of time and which data belongs to historical transaction records that are stored for a long time but accessed less frequently. At the same time, the access situation of the original message data is counted to understand the distribution of transaction messages in different protocol formats during storage and calculation. Based on these analysis results, the data classification results are obtained. According to the data classification results, the data storage structure of the local cache space is constructed to improve the access efficiency of high-frequency access data. The LRU algorithm is used for memory management and data elimination processing, that is, the data with a higher recent access frequency is stored in the local cache first, and when the cache space is insufficient, the least recently accessed data is automatically eliminated to ensure that the high-frequency data can always be retained in the memory and improve the response speed of the system. Through this mechanism, a local cache data index table is established, which enables the system to quickly locate and access cache data, reduce the delay of database queries, and improve the real-time performance of the payment transaction system. In order to optimize the distributed management of data storage, consistent hash calculation is performed on the storage nodes of the distributed cache layer to ensure that the data is evenly distributed on multiple storage nodes. Using the consistent hashing algorithm, M discrete points are mapped to the hash ring, and virtual nodes are mapped to physical storage nodes to build a node mapping relationship table. In this way, the overhead of data migration is reduced, and the cost of data redistribution is minimized when storage nodes change (such as adding or removing nodes), thereby improving the scalability and stability of the system. After the consistent hashing ring is built, the historical transaction data is sharded to ensure the balance of data storage. The hash value is calculated according to the data sharding key (such as transaction ID, user ID or timestamp), and the storage location of the data is determined on the consistent hashing ring to obtain the initial data distribution plan. Each piece of historical transaction data is automatically mapped to the corresponding storage node according to the hash calculation result, thereby achieving load-balanced storage of data and avoiding performance bottlenecks caused by excessive data concentration in specific storage nodes. In a distributed storage environment, in order to ensure that the computing resource utilization and network transmission delay of each storage node are in the optimal state, key performance indicators such as computing resource occupancy and network delay of the storage node are collected in real time, and the health of each node is calculated based on a weighted scoring model. The weighted scoring model combines multiple factors such as CPU usage, memory occupancy, disk I / O performance, and network bandwidth to assign a health score to each storage node and build a load balancing schedule that reflects the current load of different storage nodes so that optimization adjustments can be made when the storage node load is unbalanced. According to the load balancing schedule, dynamic migration of data shards is performed to optimize the distribution strategy of data storage.In order to reduce the cost of data migration, the minimum migration cost principle is adopted, that is, when redistributing data, the amount of data movement is minimized to reduce the computing and storage costs of the system. When performing data migration, storage nodes with higher health and lower load are given priority to receive some data shards, and the data shard storage scheme is updated after the migration is completed to ensure that the new data storage structure can adapt to the system load. Through this optimization strategy, efficient data storage and access capabilities are maintained in a dynamically changing transaction environment, thereby improving the computing efficiency of real-time risk control and the processing performance of multi-channel payment transactions.

[0020] In one example, multi-dimensional correlation analysis is performed on multi-channel transaction behavior flow data to obtain transaction channel correlation data, and deep feature learning is performed on the transaction channel correlation data to obtain a transaction behavior feature vector, including: Divide the time dimension data in the multi-channel transaction behavior flow data into time slots to obtain a binary feature matrix; According to the preset numerical range, the numerical data in the multi-channel transaction behavior flow data is uniquely encoded to obtain the numerical feature distribution matrix; The Euclidean distance calculation and normalization are performed on the geographical location data in the multi-channel transaction behavior flow data to obtain the location similarity matrix; Perform feature encoding on the device identification information in the multi-channel transaction behavior flow data to obtain the device fingerprint feature matrix; Perform feature contribution calculation and data association analysis on the binary feature matrix, numerical feature distribution matrix, location similarity matrix, and device fingerprint feature matrix to obtain transaction channel association data; The transaction channel association data is input into the multi-layer Transformer attention network for deep feature learning to obtain the transaction behavior feature vector.

[0021] In this example, the time dimension of multi-channel transaction behavior flow data is processed to capture the time characteristics of transaction behavior. The time slot division method is used to map the timestamp of the transaction to discrete time intervals, such as hour, half-hour or minute levels, to form a binary feature matrix. Suppose a transaction occurs at moment, and the entire time range is divided into time slots, the time slot index corresponding to the transaction is expressed as:

[0022] in, is the start time of the time series, is the width of the time slot, and Indicates the time slot index of the transaction. Construct a binary vector , where only the index is The position of is set to 1, and the rest of the positions are set to 0, forming a binary feature matrix ,in Represents a transaction sample, Represents the time slot. Process the numerical data. For continuous variables such as transaction amount and transaction frequency, the unique hot encoding method is used to discretize them, and the numerical data is mapped according to the preset numerical interval and converted into a numerical feature distribution matrix. Assume that the transaction amount The value range is , and divide it into intervals, the index of the interval where the transaction amount is located is expressed as:

[0023] in, is the width of each interval. Construct a one-hot encoding vector , where the index is The position of is set to 1, and the rest of the positions are set to 0. In transaction behavior analysis, geographic location information is also a key feature to calculate the similarity between transaction locations. and , use Euclidean distance to measure the spatial distance between two points, and the calculation formula is:

[0024] In order to ensure the balance between different data scales, the Euclidean distance is normalized so that all distance values ​​fall into the interval [0,1]. Normalization uses the minimum-maximum scaling method, that is:

[0025] in, and are the minimum and maximum values ​​of all calculated Euclidean distances. Construct a location similarity matrix , where each element Represents the normalized geographic similarity. The closer the value is to 0, the closer the transaction location is, and the closer the value is to 1, the farther the transaction location is. At the same time, the device identification information is feature encoded to identify the user's device characteristics and improve the accuracy of the risk control system. Use hash coding or embedding vectors to convert device ID, browser fingerprint, operating system version and other information into high-dimensional vector representation. For example, assuming the device ID is , through the hash function Map it to an embedding vector of fixed dimension , construct device fingerprint feature matrix . For the binary feature matrix , numerical feature distribution matrix , position similarity matrix And device fingerprint feature matrix Perform feature contribution calculation and data association analysis to construct transaction channel association data. Feature contribution calculation uses information gain or Shapley value method to measure the influence of each feature on transaction behavior prediction. For example, suppose the target variable of the transaction behavior prediction model is , then a feature The information gain is calculated as:

[0026] in, is the entropy of trading behavior, is a given feature The conditional entropy of the transaction behavior at that time. The greater the information gain, the greater the impact of the feature on the transaction behavior. Based on the association rule mining method, the correlation between different features is analyzed to obtain the transaction channel association data. The transaction channel association data is input into the multi-layer Transformer attention network for deep feature learning to obtain the transaction behavior feature vector. The Transformer network calculates the importance of the feature through the self-attention mechanism, and learns the dependency between the transaction behavior features through the multi-head attention layer to calculate the attention weight:

[0027] in, They are query matrix, key matrix and value matrix respectively. is the dimension of the key matrix. The multi-layer Transformer network extracts the global features of transaction behavior and generates a transaction behavior feature vector .

[0028] In one example, transaction channel association data is input into a multi-layer Transformer attention network for deep feature learning to obtain a transaction behavior feature vector, including: The transaction channel association data is input into the multi-head attention layer, and the dot product of the query matrix, key matrix and value matrix is ​​calculated through 8 parallel attention heads to obtain the attention weight distribution; The attention weight distribution is scaled and softmax normalized, and weighted summed with the value matrix to obtain the multi-head attention feature; The multi-head attention features are input into a two-layer feedforward neural network. The first layer uses the ReLU activation function for nonlinear transformation to obtain intermediate features, and the second layer uses linear transformation to obtain output features. Perform residual connection and layer normalization on the output features to obtain the output feature vector of the first layer Transformer block; The output feature vector of the first-layer Transformer block is sequentially input into the second-layer Transformer block and the third-layer Transformer block. Each layer of Transformer block repeatedly performs multi-head attention calculation, feedforward network transformation and normalization processing to obtain a multi-layer feature representation; Feature fusion and dimensionality reduction are performed on multi-layer feature representations to obtain the transaction behavior feature vector.

[0029] In this example, the input transaction channel association data is structured to adapt to the calculation method of the multi-head attention mechanism. Suppose the transaction channel association data matrix is ,in The dimension is ,in represents the number of transaction samples, represents the feature dimension of each transaction sample. In order to calculate the attention weight, Transformed into query matrix , key matrix Sum Matrix , through three trainable weight matrices , and It is completed by linear transformation, and the calculation formula is:

[0030] in, The dimensions of ,in is the dimension of a single attention head. In order to implement the multi-head attention mechanism, 8 independent attention heads are used, each with its own Transformation is performed so that the calculation is performed in multiple subspaces to improve the model's ability to extract features of transaction behavior. Calculate the query matrix and key matrix The dot product of To maintain numerical stability, the unnormalized attention weight distribution is obtained:

[0031] in, The shape is , indicating the importance of each transaction sample to other samples. Softmax normalization is applied to ensure that the sum of attention weights is 1:

[0032] in, represents the normalized attention weight matrix. With value matrix Perform weighted summation to generate multi-head attention features:

[0033] in, The shape is still For 8 attention heads, calculate , and then concatenate them to form the multi-head attention output:

[0034] in, is used to map back to the original feature dimension The trainable weight matrix of . The multi-head attention feature Input to a two-layer feedforward neural network to extract deep features of trading behavior. In the first layer of the feedforward network, the ReLU activation function is used for nonlinear transformation:

[0035] in, The dimension is , is the hidden layer dimension of the feedforward network, is the bias term. In the second layer of the feedforward network, a linear transformation is used to obtain the final output features:

[0036] in, The dimension is , is the bias term, Still maintain In order to stabilize the training and improve the performance of the model, residual connections and layer normalization are introduced in the calculation process of the Transformer block. The residual connection ensures that the gradient propagation between the input and output is more stable, and its calculation formula is:

[0037] Application layer normalization:

[0038] Get the output feature vector of the first layer Transformer block , which contains the deep features of transaction behavior. It is input into the second and third layers of Transformer blocks in turn. In each layer, multi-head attention calculation, feedforward network transformation and normalization are repeatedly performed to further extract the hierarchical features of transaction behavior. For the calculation of the second layer Transformer block:

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045] Similarly, the third layer Transformer block continues processing , and finally obtain a complete multi-layer feature representation of transaction behavior. In order to reduce the computational complexity and generate the transaction behavior feature vector, the multi-layer feature representation is subjected to feature fusion and dimensionality reduction. The fully connected layer is used for dimensionality reduction:

[0046] in, The dimension is , It is the dimension of the final transaction behavior feature vector. As a transaction behavior feature vector, it is used for real-time risk control decisions, enabling the payment system to efficiently identify abnormal transactions and improve the security and accuracy of payment transactions.

[0047] In one example, protocol conversion and field mapping are performed on original message data from multiple channels to obtain standard payment transaction data, including: Perform protocol type identification on multi-channel original message data to obtain a protocol type identification set, and parse and split the multi-channel original message data according to the protocol type identification set to obtain an original field data set; The original field data set is input into the field mapping rule library for matching to obtain a standard field mapping table, and the original field data set is converted according to the standard field mapping table to obtain initial standard data; The initial standard data is subjected to data integrity check and consistency check to obtain verification result data, and the initial standard data is subjected to data completion and cleaning processing based on the verification result data to obtain standard payment transaction data.

[0048] In this example, protocol type analysis is performed on original message data from multiple channels to determine the communication protocol format used by transaction data from different sources. ,in Indicates The original message of a transaction, while the transaction data of different channels uses different protocol formats, such as ISO 8583, XML, JSON or fixed-length messages. Build a protocol identification model , to automatically determine the protocol type based on the structure and field characteristics of the message:

[0049] in, Indicates The protocol type of the transaction, get the protocol type identifier set , used to guide the subsequent parsing process. According to the protocol type identification set, the original message data of multiple channels is parsed and split to extract the transaction-related field information. For structured protocols (such as JSON, XML), the key-value pair fields are extracted based on the hierarchical parser:

[0050] in, Indicates For fixed-length formats (such as ISO 8583 messages), the fields are split based on predefined offsets:

[0051] in, and Respectively represent The start and end offsets of the fields to get the complete original field data set . Perform field standardization on the original field data to ensure that fields in different protocol formats are mapped to a unified data structure. Build a field mapping rule library , which stores the corresponding relationship between fields in different protocol formats, for example:

[0052] The original field dataset Enter the field mapping rule base for matching to generate a standard field mapping table:

[0053] in, Indicates The standard field mapping table of the transaction is used to convert the original field data set to obtain the initial standard data:

[0054] in, Represents the converted standardized transaction data. Perform data integrity and consistency check on the standardized transaction data. The purpose of integrity check is to check whether there are any missing key fields. For each transaction, define the integrity check function:

[0055] in, is the index collection of all key fields. To indicate the function, when a key field is empty, , indicating that the integrity check failed. The goal of consistency check is to check whether the field value conforms to the expected rules. Suppose the consistency check function is:

[0056] in, Represents the format validation rules for a specific field. If all fields meet the format requirements, ,otherwise The comprehensive results of the integrity and consistency check are:

[0057] when , it means the data has passed the verification, otherwise data needs to be completed and cleaned. For data that has not passed the verification, historical data or business rules are used to complete it. For example, if the transaction time is missing, the time information embedded in the transaction serial number is used to fill it. After cleaning the abnormal data, such as removing non-standard characters, normalizing the currency list and cleaning, standard payment transaction data is obtained.

[0058] In one example, a multi-channel transaction processing flow is subjected to random network modeling based on standard payment transaction data to obtain processing delay threshold parameters, including: According to the data preprocessing module, feature extraction module, risk control decision module and response generation module, the processing node information in the standard payment transaction data is hierarchically divided and multi-level service nodes are constructed to obtain the processing node topology structure; According to the processing node topology, the data flow relationship between nodes is modeled by Markov chain and the state transfer matrix is ​​constructed to obtain the node dependency graph; Perform maximum likelihood estimation and service time distribution parameter fitting on the historical processing data of each node in the processing node topology structure to obtain the node service time probability density function; The node service time probability density function is Laplace transformed and the cumulative moment of the service time distribution is calculated to obtain the end-to-end delay distribution function, and the probability integral of the end-to-end delay distribution function is calculated according to the preset service quality requirements to obtain the delay constraint condition; According to the delay constraint conditions and the node dependency graph, numerical solution and upper bound calculation of the optimal processing delay are performed to obtain the processing delay threshold parameter.

[0059] In this example, the payment transaction data processing flow is structured and modeled to analyze the execution logic and data flow of each processing node. , where each Represents the complete data of a transaction, including user identity information, transaction amount, transaction time, payment channel, device information, etc. In the actual operation of the payment system, each transaction goes through multiple processing links, including data preprocessing, feature extraction, risk control decision-making and final response generation. These modules are used as the main processing layer, and their internal nodes are refined to build a multi-level service node structure. In order to establish the processing node topology, the processing system is assumed to include A different set of processing nodes , where each node Corresponding to a specific computing task, such as data analysis, risk assessment, transaction confirmation, etc. In order to build the connection relationship between nodes, define the data flow function To represent the node To Node The data flow relationship between them is:

[0060] Through this step, the processing flow of the entire payment transaction system is represented as a directed graph ,in is the set of all processing nodes, and the edge set Satisfied by all Node pair The data flow relationship between nodes is modeled according to the processing node topology structure, so as to analyze the flow characteristics of transactions between different processing nodes. Markov chain is used to describe the state transition process of data flow. Assume that the state set in the system is , and define the state transfer matrix :

[0061] in, Indicates transaction data from the node Flow to The probability of is estimated through historical data. Assume that in the historical transaction records, there are Secondary data stream from node Transfer to , then the state transition probability is calculated as:

[0062] Finally, the state transfer matrix is ​​obtained , which is used to describe the data flow pattern in the payment system and build a node dependency graph. The service time distribution of each node in the processing node topology is modeled to calculate the system's delay characteristics. Assume that each node The time to process a transaction follows a probability distribution , and use the maximum likelihood estimation method to fit its parameters. For Node The goal of maximum likelihood estimation is to find the optimal distribution parameter , so that:

[0063] Common distribution fitting methods include exponential distribution, normal distribution, or gamma distribution. For example, if the processing time of a node follows an exponential distribution:

[0064] in, Representation Node The probability density at ; Is the processing rate parameter, which means the number of nodes per unit time The probability of completing a transaction; the characteristic of exponential distribution is that it is memoryless, that is, the processing time of the current transaction will not affect the processing time of future transactions. , using the "maximum likelihood estimation" method. Let the node The historical processing time sample set is:

[0065] in, Indicates Transactions at the node The optimal parameter of the exponential distribution is Calculated as:

[0066] If historical data indicates that processing time is more normally distributed:

[0067] in, represents the average processing time, i.e., the node The average time to process a transaction; represents the variance of processing time, which describes the fluctuation degree of node processing time. Then the parameter estimation is:

[0068] Obtain the service time probability density function of all nodes. After obtaining the node service time distribution, calculate the end-to-end delay distribution of the entire system. Perform Laplace transform on the service time probability density function of each node:

[0069] The Laplace transform of the end-to-end delay is expressed as:

[0070] right Perform inverse Laplace transform to obtain the delay distribution function of the trading system , and calculate the cumulative distribution function:

[0071] in, Indicates the maximum allowed delay; if Meet service quality requirements (such as 99% of transactions in If the system performance is acceptable, otherwise it needs to be optimized. According to the delay constraint and the node dependency graph, the optimal processing delay upper bound is calculated. Suppose the optimal delay allocation for each node is , then the optimization objective is expressed as:

[0072] subject to

[0073]

[0074] After solving the optimization problem, the processing delay threshold parameters are obtained:

[0075] This value represents the maximum transaction processing time of the entire system under optimal conditions and is used to dynamically adjust computing resources and optimize the response speed of the payment system.

[0076] In one example, the state space and reward function of the TR-A2C deep reinforcement learning network are constructed based on the transaction behavior feature vector and the processing delay threshold parameter, and a multi-channel real-time risk control decision strategy is obtained through parallel strategy optimization, including: The transaction behavior feature vector is divided into feature subspaces according to transaction value, transaction frequency, transaction location and transaction device, and a delay monitoring vector is constructed based on the processing delay threshold parameter. The state space embedding vector is obtained by multi-dimensional tensor splicing. The state space embedding vector is subjected to sparsity constraints and regularization to obtain a standardized state space, and the standardized state space is mapped to the risk control action space to obtain an action space set; Based on the standardized state space, the Transformer encoding layer, policy layer and value assessment layer are connected in series to obtain the TR-A2C deep reinforcement learning network. The TR-A2C deep reinforcement learning network is sampled in parallel, and the interaction data is collected synchronously in multiple environments to obtain the policy trajectory sequence dataset. Calculate the weighted combination of delay loss and risk control loss based on the strategy trajectory sequence data set, and calculate the advantage function based on the weighted combination to obtain the strategy gradient; The policy gradient is input into the Adam optimizer for parameter update and loop iteration to obtain a multi-channel real-time risk control decision-making strategy.

[0077] In this example, the transaction behavior feature vector is reasonably divided into feature subspaces so that the model can learn different dimensional information of transaction data and make more accurate risk control decisions in the reinforcement learning process. Suppose the transaction behavior feature vector is:

[0078] in, Indicates The feature vector of the transaction, is the total number of transaction samples. In order to make full use of the information of transaction data, it is divided according to different feature categories of transactions to obtain four feature subspaces: transaction numerical feature subspace , including numerical information such as transaction amount and payment method; transaction frequency feature subspace , including the number of recent transactions of users, average transaction interval, etc.; transaction geographic location feature subspace , including the latitude and longitude of the transaction, transaction distance, etc.; transaction equipment feature subspace , including device fingerprint, operating system information, etc. Respectively represent the dimensions of each feature subspace, then it can be expressed as:

[0079] At the same time, in order to measure the transaction processing delay, a delay monitoring vector is constructed based on the processing delay threshold parameter. ,in Each element of Representative The processing delay of a transaction:

[0080] in, The actual processing delay of the transaction Normalized to get:

[0081] in, and Respectively represent the minimum and maximum delays observed in the trading system. Delay monitoring vector Perform multi-dimensional tensor concatenation to construct a state space embedding vector:

[0082] in, As the input state vector of reinforcement learning. The state space embedding vector is sparsely constrained and regularized. Define the L1 regularization term:

[0083] in, is the regularization coefficient, Represents the L1 norm of the state vector, which is used to encourage the sparsity of features so that the weight of irrelevant information tends to zero. Perform normalization operation:

[0084] in, and They are After the above processing, the standardized state space is obtained. , and map it to the risk control action space, that is, define the risk control decision set:

[0085] in, Represents the number of risk control strategies. Based on the standardized state space Design the TR-A2C deep reinforcement learning network, which consists of an encoding layer, a policy layer, and a value assessment layer. The encoding layer uses a self-attention mechanism to process transaction status information:

[0086] in, They are query matrix, key matrix and value matrix respectively. is the dimension of the attention head. The probability distribution of risk control decisions generated by the strategy layer:

[0087] in, It is a neural network used to calculate the risk control strategy score. At the same time, the value assessment layer calculates the state value function:

[0088] To train the TR-A2C network, parallel environment sampling is performed to collect reinforcement learning trajectories in multiple trading environments simultaneously:

[0089] in, is the reward for trading decisions. Defining delay loss and risk control losses :

[0090]

[0091] in, is the target latency threshold, is the current transaction processing delay. Calculate the weighted combination loss:

[0092] in, and is the weight hyperparameter. Calculate the advantage function To update the policy:

[0093] in, is the advantage function, measuring the current action Advantages: Is the discount factor, which is used to measure the impact of future rewards on current decisions. The value range is ; is the expected value of the next state; is the expected value of the current state; is an instant reward, indicating the risk control benefit of the current transaction. , indicating that the currently selected risk control strategy is better than the average strategy; if , indicating that the current action's benefit is lower than the average, and the strategy needs to be adjusted. Calculate the strategy gradient:

[0094] in, is the policy objective function, which represents the policy Expected return; are the parameters of the strategy network, i.e., the weights of the neural network used to optimize risk control decisions; It's about parameters The gradient of is used to update the policy network to increase the probability of high-reward actions; represents the expectation for all transaction samples; is the action probability distribution of the current policy network, indicating that in state Next Select The probability of Calculate the log probability of the action; is the advantage function, which is used to measure the direction of improvement of the current strategy. Use the optimizer to perform parameter updates:

[0095] in, are the parameters of the policy network; is the learning rate, which controls the step size of gradient update; It is a policy gradient, which is used to optimize the policy network to make it more inclined to high-yield decisions. After multiple rounds of iterations, the optimal multi-channel real-time risk control decision strategy is finally obtained, which automatically selects risk control actions according to the transaction status, realizes intelligent interception of fraudulent transactions, and improves the security of the payment system and the transaction experience.

[0096] Among them, the policy gradient is input into the Adam optimizer for parameter update and cyclic iteration to obtain a multi-channel real-time risk control decision strategy, including: constructing a time-varying Riccati equation for the convergence rate and decision amplitude constraints of the policy gradient, and designing a time-varying parameter scheduler according to the system state error and the decision output boundary to obtain a parameter update matrix; based on the parameter update matrix, the policy layer and value evaluation layer of the TR-A2C deep reinforcement learning network are adjusted online to obtain a time-varying gain sequence; the time-varying gain sequence is input into the momentum term and second-order moment estimation term of the Adam optimizer to perform adaptive learning rate calculation to obtain the optimizer parameter group; according to the optimization The policy gradient is normalized and amplitude-limited by the optimizer parameter group, and the constrained gradient vector is obtained through nonlinear projection; the constrained gradient vector is sampled according to the preset batch size, and the sample weight is calculated to obtain a weighted training data set; small-batch stochastic gradient descent is performed on the weighted training data set, and the convergence time is estimated through linear regression to obtain a time-varying learning rate sequence; the update step size of the optimizer is dynamically adjusted according to the time-varying learning rate sequence, and the early stopping condition is set for convergence judgment to obtain the optimal parameter solution; based on the optimal parameter solution, a risk control decision function is constructed, and the convergence boundary is determined through finite-time stability analysis to obtain a multi-channel real-time risk control decision strategy.

[0097] In one example, the intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction also includes: Perform access frequency statistics and data volume analysis on multi-channel transaction behavior flow data and multi-channel original message data to obtain data classification results; The data storage structure of the local cache space is constructed according to the data classification result, and the memory management and data elimination processing are performed through the LRU algorithm to obtain the local cache data index table; Perform consistent hash calculation on the storage nodes of the distributed cache layer to generate M discrete points to form a hash ring, and map the virtual nodes to the physical nodes to obtain a node mapping relationship table; Calculate the hash value of historical transaction data according to the data sharding key, determine the data storage location on the consistent hash ring, and obtain the initial data distribution plan; The computing resource occupancy rate and network transmission delay of each storage node are collected in real time, and the node health is calculated based on the weighted scoring model to obtain the load balancing schedule; Perform dynamic migration of data shards according to the load balancing schedule, reallocate data storage locations according to the principle of minimum migration cost, and obtain a multi-channel data shard storage solution.

[0098] In this example, the data access pattern is analyzed to develop the optimal storage and caching strategy. Suppose the transaction behavior flow data set is:

[0099] in, Indicates A complete record of a transaction, including transaction time, transaction amount, transaction equipment, transaction channel and other information. For the original message data set, let it be:

[0100] in, Represents the original messages from different payment channels. Statistics are collected on the access frequency and data volume of the above data. Assume that the access frequency function is:

[0101] in, Indicates transaction In the time window Similarly, for data volume analysis, define the data size function:

[0102] in, Representative transaction No. fields, is the total number of fields, is the field size. By calculating the and , perform data classification. According to the data classification results, build a local cache structure for high-frequency data to improve query efficiency and reduce database pressure. Suppose the cache space size is , then the cache data set is:

[0103] in, is the minimum access frequency required by the cache. In order to optimize cache utilization, the LRU algorithm is used for memory management, that is, when the cache space reaches the upper limit, the least recently used data is removed. Suppose the last access time of the cached data is , then the LRU elimination strategy is defined as:

[0104] That is, remove the earliest unaccessed data and cache the data index table. Maintain the latest cache records in:

[0105] After optimizing the local cache, the distributed storage layer is optimized, consistent hashing is performed, and a hash ring is generated to ensure that data is evenly distributed on the storage nodes. Suppose the storage node set is:

[0106] Through hash function Calculate the position of each node:

[0107] and will Hash values ​​are mapped to the hash ring to minimize data migration costs. Introducing virtual nodes , where each physical node Mapped to multiple virtual nodes:

[0108] This effectively reduces the data skew problem and forms a node mapping relationship table:

[0109] Calculate the hash value according to the data sharding key to determine the storage location of the transaction data. Suppose the key field set of the transaction data is , define the data shard key:

[0110] Then find the nearest storage node on the hash ring:

[0111] Ensure that each transaction data is evenly distributed to the appropriate storage node and form an initial data distribution plan. During the data storage operation, monitor the computing resource utilization and network transmission delay of the storage node in real time to evaluate the node health. The CPU usage is , the memory usage is , the disk IO rate is , the network delay is , then the node health is calculated as:

[0112] in, is a weight parameter that reflects the impact of different resources on system performance. After calculating the health of all storage nodes, a load balancing schedule is generated:

[0113] And readjust the data storage according to the load balancing strategy. In order to reduce the cost of data migration, the principle of minimum migration cost is adopted to perform dynamic migration of data shards. Load If the threshold is exceeded, some data will be migrated to nodes with higher health. . Define the migration cost:

[0114] in, Indicates the amount of migrated data; Indicates the additional delay caused by migration; and is the weight parameter. The optimal migration target node is:

[0115] Through this step, the data will be migrated to the target node with the lowest cost, and finally a multi-channel data sharding storage solution will be formed.

[0116] Reference Figure 2 This embodiment provides an intelligent bank payment and settlement device that supports real-time risk control and multi-channel interaction, including: The correlation analysis unit 1 is used to perform multi-dimensional correlation analysis on the multi-channel transaction behavior flow data to obtain transaction channel correlation data, and perform deep feature learning on the transaction channel correlation data to obtain a transaction behavior feature vector; Conversion unit 2, used for performing protocol conversion and field mapping processing on multi-channel original message data to obtain standard payment transaction data; Modeling and analysis unit 3, used to perform random network modeling on the multi-channel transaction processing flow according to standard payment transaction data to obtain a processing delay threshold parameter; The risk control decision unit 4 is used to construct the state space and reward function of the TR-A2C deep reinforcement learning network according to the transaction behavior feature vector and the processing delay threshold parameter, and obtain the multi-channel real-time risk control decision strategy through parallel strategy optimization.

[0117] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0118] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0119] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0120] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0121] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0122] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0123] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction, characterized in that: The following steps are involved: Performing multi-dimensional correlation analysis on multi-channel transaction behavior flow data to obtain transaction channel correlation data, and performing deep feature learning on the transaction channel correlation data to obtain a transaction behavior feature vector; Perform protocol conversion and field mapping on original message data from multiple channels to obtain standard payment transaction data; Performing random network modeling on the multi-channel transaction processing flow according to the standard payment transaction data to obtain a processing delay threshold parameter; The state space and reward function of the TR-A2C deep reinforcement learning network are constructed according to the transaction behavior feature vector and the processing delay threshold parameter, and a multi-channel real-time risk control decision strategy is obtained through parallel strategy optimization.

2. The intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction according to claim 1, characterized in that: The multi-dimensional correlation analysis is performed on the multi-channel transaction behavior flow data to obtain transaction channel correlation data, and deep feature learning is performed on the transaction channel correlation data to obtain a transaction behavior feature vector, including: Divide the time dimension data in the multi-channel transaction behavior flow data into time slots to obtain a binary feature matrix; According to the preset numerical range, the numerical data in the multi-channel transaction behavior flow data is uniquely encoded to obtain the numerical feature distribution matrix; The Euclidean distance calculation and normalization are performed on the geographical location data in the multi-channel transaction behavior flow data to obtain the location similarity matrix; Perform feature encoding on the device identification information in the multi-channel transaction behavior flow data to obtain the device fingerprint feature matrix; Performing feature contribution calculation and data association analysis on the binary feature matrix, the numerical feature distribution matrix, the position similarity matrix, and the device fingerprint feature matrix to obtain transaction channel association data; The transaction channel association data is input into a multi-layer Transformer attention network for deep feature learning to obtain a transaction behavior feature vector.

3. The intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction according to claim 2 is characterized in that: The step of inputting the transaction channel association data into a multi-layer Transformer attention network for deep feature learning to obtain a transaction behavior feature vector includes: Input the transaction channel association data into the multi-head attention layer, and calculate the dot product of the query matrix, the key matrix and the value matrix through 8 parallel attention heads to obtain the attention weight distribution; The attention weight distribution is scaled and softmax normalized, and a weighted sum operation is performed with the value matrix to obtain a multi-head attention feature; The multi-head attention features are input into a two-layer feedforward neural network, the first layer uses a ReLU activation function to perform nonlinear transformation to obtain intermediate features, and the second layer uses a linear transformation to obtain output features; Performing residual connection and layer normalization processing on the output features to obtain an output feature vector of the first layer Transformer block; The output feature vector of the first-layer Transformer block is sequentially input into the second-layer Transformer block and the third-layer Transformer block. Each layer of Transformer block repeatedly performs multi-head attention calculation, feedforward network transformation and normalization processing to obtain a multi-layer feature representation; The multi-layer feature representation is subjected to feature fusion and dimensionality reduction processing to obtain a transaction behavior feature vector.

4. The intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction according to claim 1, characterized in that: The protocol conversion and field mapping processing of the original message data from multiple channels to obtain standard payment transaction data includes: Performing protocol type identification on the original message data of multiple channels to obtain a protocol type identification set, and parsing and splitting the original message data of multiple channels according to the protocol type identification set to obtain an original field data set; Input the original field data set into the field mapping rule library for matching to obtain a standard field mapping table, and perform field conversion on the original field data set according to the standard field mapping table to obtain initial standard data; The initial standard data is subjected to data integrity check and consistency check to obtain verification result data, and the initial standard data is subjected to data completion and cleaning processing according to the verification result data to obtain standard payment transaction data.

5. The intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction according to claim 1, characterized in that: The performing random network modeling on the multi-channel transaction processing flow according to the standard payment transaction data to obtain a processing delay threshold parameter includes: According to the data preprocessing module, the feature extraction module, the risk control decision module and the response generation module, the processing node information in the standard payment transaction data is hierarchically divided and a multi-level service node is constructed to obtain a processing node topology structure; According to the processing node topology structure, Markov chain modeling and state transfer matrix construction are performed on the data flow relationship between nodes to obtain a node dependency graph; Performing maximum likelihood estimation and service time distribution parameter fitting on the historical processing data of each node in the processing node topology structure to obtain a node service time probability density function; Performing Laplace transformation on the node service time probability density function and calculating the cumulative moment of the service time distribution to obtain an end-to-end delay distribution function, and performing probability integral calculation on the end-to-end delay distribution function according to a preset service quality requirement to obtain a delay constraint condition; Numerical solution and upper bound calculation of optimal processing delay are performed according to the delay constraint condition and the node dependency graph to obtain a processing delay threshold parameter.

6. The intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction according to claim 1, characterized in that: The state space and reward function of the TR-A2C deep reinforcement learning network are constructed according to the transaction behavior feature vector and the processing delay threshold parameter, and a multi-channel real-time risk control decision strategy is obtained through parallel strategy optimization, including: The transaction behavior feature vector is divided into feature subspaces according to transaction value, transaction frequency, transaction geographic location and transaction device, and a delay monitoring vector is constructed based on the processing delay threshold parameter, and a state space embedding vector is obtained by multi-dimensional tensor splicing; Performing sparsity constraint and regularization processing on the state space embedding vector to obtain a standardized state space, and performing risk control action space mapping on the standardized state space to obtain an action space set; Based on the standardized state space, the Transformer encoding layer, the strategy layer and the value assessment layer are connected in series to obtain a TR-A2C deep reinforcement learning network, and the TR-A2C deep reinforcement learning network is sampled in parallel, and interaction data is collected synchronously in multiple environments to obtain a strategy trajectory sequence data set; Calculating a weighted combination of delay loss and risk control loss according to the strategy trajectory sequence data set, and calculating an advantage function based on the weighted combination to obtain a strategy gradient; The policy gradient is input into the Adam optimizer for parameter update and loop iteration to obtain a multi-channel real-time risk control decision-making strategy.

7. The intelligent bank payment and settlement method supporting real-time risk control and multi-channel interaction according to claim 1, characterized in that: The intelligent bank payment settlement method supporting real-time risk control and multi-channel interaction also includes: Performing access frequency statistics and data volume analysis on the multi-channel transaction behavior flow data and the multi-channel original message data to obtain data classification results; The data storage structure of the local cache space is constructed according to the data classification result, and the memory management and data elimination processing are performed through the LRU algorithm to obtain the local cache data index table; Perform consistent hash calculation on the storage nodes of the distributed cache layer to generate M discrete points to form a hash ring, and map the virtual nodes to the physical nodes to obtain a node mapping relationship table; Calculate the hash value of historical transaction data according to the data sharding key, determine the data storage location on the consistent hash ring, and obtain the initial data distribution plan; The computing resource occupancy rate and network transmission delay of each storage node are collected in real time, and the node health is calculated based on the weighted scoring model to obtain the load balancing schedule; Perform dynamic migration of data shards according to the load balancing schedule, reallocate data storage locations according to the principle of minimum migration cost, and obtain a multi-channel data shard storage solution.

8. An intelligent bank payment and settlement device supporting real-time risk control and multi-channel interaction, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the device comprises: A correlation analysis unit, used to perform multi-dimensional correlation analysis on multi-channel transaction behavior flow data to obtain transaction channel correlation data, and perform deep feature learning on the transaction channel correlation data to obtain a transaction behavior feature vector; A conversion unit, used to perform protocol conversion and field mapping processing on multi-channel original message data to obtain standard payment transaction data; A modeling and analysis unit, configured to perform random network modeling on a multi-channel transaction processing flow according to the standard payment transaction data to obtain a processing delay threshold parameter; The risk control decision unit is used to construct the state space and reward function of the TR-A2C deep reinforcement learning network according to the transaction behavior feature vector and the processing delay threshold parameter, and obtain a multi-channel real-time risk control decision strategy through parallel strategy optimization.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images