Intelligent payment routing and clearing settlement optimization method and system using deep reinforcement learning

Abstract

The embodiment of the application provides a kind of intelligent payment routing and clearing settlement optimization method and system using deep reinforcement learning, it is related to deep reinforcement learning technical field, first, the real-time corresponding relationship of payment request key information and clearing settlement execution rule is established, to provide basis for subsequent decision-making;Then build the deep reinforcement learning collaborative evolution model containing information perception, interactive simulation, decision output and evolution update node;The interactive relationship of payment channel data and clearing settlement rules is analyzed by the model, and the payment routing dynamic adjustment path and the clearing settlement process adaptation strategy are generated;According to the strategy, payment request transmission and clearing settlement operation are guided;Finally, relevant data is collected to update the model decision parameter.The application improves the success rate, efficiency, stability, flexibility and intelligent level of payment system.

Description

Technical Field

[0001] This application relates to the field of deep reinforcement learning technology, and more specifically, to an intelligent payment routing and clearing optimization method and system utilizing deep reinforcement learning. Background Technology

[0002] With the rapid development of e-commerce and mobile payments, the number of payment requests has exploded, and payment scenarios have become increasingly complex and diverse. Traditional payment routing and clearing methods have gradually revealed many problems when dealing with large-scale, high-concurrency payment requests.

[0003] Existing payment routing methods typically select payment channels based on preset rules and fixed priorities, lacking the ability to dynamically perceive and analyze the real-time payment environment and payment request characteristics. This results in an inability to flexibly adjust payment routes when faced with network congestion, payment channel failures, or differences in rates among different payment channels, easily leading to problems such as payment failures, payment delays, or increased payment costs.

[0004] In terms of clearing and settlement, traditional clearing and settlement processes are often fixed and standardized, making it difficult to dynamically adapt to different payment requests and business scenarios. For example, some special payment transactions, such as cross-border payments and large-value payments, may require different clearing and settlement rules and processes, but existing clearing and settlement systems cannot automatically identify and make corresponding adjustments, thus affecting the efficiency and accuracy of clearing and settlement.

[0005] Furthermore, existing payment routing and clearing systems are mostly independent, lacking effective coordination mechanisms. This prevents the optimal matching between payment routing selection and clearing execution during the payment process, hindering the full utilization of the overall performance of the payment system. Summary of the Invention

[0006] In view of this, the purpose of this application is to provide an intelligent payment routing and clearing optimization method and system that utilizes deep reinforcement learning.

[0007] According to a first aspect of this application, a smart payment routing and clearing optimization method utilizing deep reinforcement learning is provided, the method comprising: Establish a real-time correspondence between key information of payment requests and clearing and settlement execution rules. The real-time correspondence is used to reflect the clearing and settlement operation standards and execution requirements corresponding to various key information of payment requests. A deep reinforcement learning co-evolution model is constructed, which includes information perception nodes, interactive simulation nodes, decision output nodes, and evolution update nodes. By analyzing the interaction between payment channel transmission data and clearing and settlement execution rules through a deep reinforcement learning collaborative evolution model, dynamic adjustment paths for payment routing and adaptation strategies for clearing and settlement processes are generated. Based on the payment routing, the path is dynamically adjusted to guide the payment request to complete the data transmission within the corresponding payment channel, and the entire clearing and settlement process is executed synchronously according to the clearing and settlement process adaptation strategy. The system collects transmission data of the payment routing dynamic adjustment path and execution data of the clearing and settlement process adaptation strategy, and transmits the transmission data of the payment routing dynamic adjustment path and execution data of the clearing and settlement process adaptation strategy to the evolution update node of the deep reinforcement learning co-evolution model to update the decision parameters of the deep reinforcement learning co-evolution model.

[0008] According to a second aspect of this application, a smart payment routing and clearing optimization system utilizing deep reinforcement learning is provided. The smart payment routing and clearing optimization system utilizing deep reinforcement learning includes a processor and a readable storage medium. The readable storage medium stores a program that, when executed by the processor, implements the aforementioned smart payment routing and clearing optimization method utilizing deep reinforcement learning.

[0009] Based on any of the above aspects, by establishing a real-time correspondence between key information in payment requests and clearing and settlement execution rules, the operational standards and execution requirements for clearing and settlement under different payment requests can be determined. The constructed deep reinforcement learning collaborative evolution model integrates information perception, interactive simulation, decision output, and evolution update functions. It can deeply analyze the complex interaction between payment channel transmission data and clearing and settlement execution rules, thereby generating highly accurate and dynamically adaptable payment routing adjustment paths and clearing and settlement process adaptation strategies. Guiding payment request transmission and clearing and settlement operations based on these strategies effectively avoids problems such as payment failures, delays, and increased costs caused by fixed rules in traditional methods, significantly improving the success rate and efficiency of the payment system. Simultaneously, collecting transmission and execution data and feeding it back to the evolution update node continuously optimizes the model's decision parameters, enabling the system to continuously adapt to the ever-changing payment environment and business needs, further enhancing the stability, flexibility, and intelligence of the payment system. Attached Figure Description

[0010] Figure 1 A flowchart illustrating the intelligent payment routing and clearing optimization method utilizing deep reinforcement learning provided in an embodiment of this application is shown. Figure 2 This paper illustrates a schematic diagram of the component structure of the intelligent payment routing and clearing optimization system utilizing deep reinforcement learning provided in an embodiment of this application. Detailed Implementation

[0011] Figure 1A flowchart illustrating the intelligent payment routing and clearing optimization method utilizing deep reinforcement learning provided in this application embodiment is shown, with detailed steps described below.

[0012] Step S110: Establish a real-time correspondence between key information of payment requests and clearing and settlement execution rules. The real-time correspondence is used to reflect the clearing and settlement operation standards and execution requirements corresponding to various key information of payment requests.

[0013] In this embodiment, the key information of the payment request includes a business type identifier, payment amount range, initiator account type, recipient type, and timeliness requirement level. This key information exists as structured fields in the header and payload areas of the payment request data packet. The clearing and settlement execution rules include the scope of fund collection, account record field requirements, fund transfer timeliness threshold, and voucher generation elements. These rules are stored in the rule base as entries, with each rule entry containing a precondition field and a rule content field. When establishing a real-time correspondence, the key information fields in the payment request are first read. The content of each field is compared one by one with the precondition fields of each rule entry in the rule base. When all the field contents of the payment request's key information match the precondition field contents of a certain rule entry, the rule content of that rule entry is associated with the payment request's key information, generating a real-time correspondence record. The real-time correspondence record includes a payment request identifier, a set of key information fields, the corresponding clearing and settlement execution rule identifier, and a rule content summary. After the record is generated, it is synchronously stored in a memory cache and a persistent database. The memory cache is used for subsequent millisecond-level queries, and the persistent database is used for tracing and auditing the rule association history.

[0014] Step S120: Construct a deep reinforcement learning co-evolution model, which includes information perception nodes, interactive simulation nodes, decision output nodes, and evolution update nodes.

[0015] In this embodiment, each node of the deep reinforcement learning collaborative evolution model is an independent functional unit, and data interaction is achieved through a standardized data interface. The information perception node receives external input data related to payment channel transmission and clearing and settlement execution rules, and verifies the format of the input data, including field integrity and data type consistency. After successful verification, the data is forwarded to the interactive simulation node. The interactive simulation node receives the data forwarded by the information perception node, simulates the interaction process between payment channel transmission behavior and clearing and settlement execution rule processing behavior, and records state changes and result feedback during the interaction process. The decision output node receives the output data from the interactive simulation node and generates payment routing dynamically, adjusting paths and clearing and settlement process adaptation strategies through reinforcement learning logic. The evolution update node receives feedback data generated by the model's operation and iteratively updates the model's decision parameters. The functional boundaries and data interaction formats of each node are solidified in the model construction phase through configuration files, which include node names, input data field lists, output data field lists, and data transmission protocols.

[0016] Step S121: Collect historical interaction data between the payment channel and the clearing and settlement process. The historical interaction data between the payment channel and the clearing and settlement process includes the transmission response data, resource usage data and process connection data of the payment channel under different clearing and settlement execution rules.

[0017] In this embodiment, historical interaction data is collected through a log collection component deployed at the interface between the payment system and the clearing and settlement system. The log collection component scans the system log files at preset time intervals. Transmission response data is extracted from the payment channel's transmission logs, including the start time, end time, and response status identifier of the payment request transmission within the payment channel. Resource usage data is extracted from the payment channel's resource monitoring logs, including the percentage of network bandwidth used, server CPU usage, and memory usage during transmission. Process connection data is extracted from the interaction logs between the payment system and the clearing and settlement system, including the time difference between the payment channel's transmission completion time and the clearing and settlement process start time, the clearing and settlement process's confirmation of receipt of the data transmitted by the payment channel, and the clearing and settlement process's processing result identifier based on the data transmitted by the payment channel. The collected historical interaction data is stored in time-series format, with a storage structure including a timestamp field, a payment channel identifier field, a clearing and settlement execution rule identifier field, a data type field, and a data content field.

[0018] Step S122: Based on the historical interaction data, extract descriptive information of the payment channel transmission-related data. The descriptive information includes data transmission efficiency data and transmission stability data characterizing transmission performance, as well as process compatibility evaluation data and rule adaptability evaluation data characterizing interaction features, which together constitute a set of descriptive data related to payment channel transmission. Based on the historical interaction data, extract execution detail information of the clearing and settlement execution rules. The execution detail information includes data describing the fund processing status, account record status, fund transfer status, and voucher generation status at different processing stages, which together construct a set of details for the clearing and settlement execution rules.

[0019] In this embodiment, when extracting descriptive information about payment channel transmission-related data, the system first reads transmission response data from historical interaction data, calculates the ratio of the number of payment request transmissions to the total transmission time under the same clearing and settlement execution rules for the same payment channel, and obtains data transmission efficiency data. It then calculates the ratio of the number of successful transmission requests to the total number of transmission requests under the same clearing and settlement execution rules for the same payment channel, and obtains transmission stability data. Next, it reads process connection data from historical interaction data, calculates the ratio of the number of requests whose transmission data from the same payment channel matches the clearing and settlement process data format to the total number of requests, and obtains process compatibility assessment data. Finally, it calculates the ratio of the number of requests whose transmission behavior conforms to the clearing and settlement execution rules to the total number of requests, and obtains rule adaptability assessment data. The aforementioned data transmission efficiency data, transmission stability data, process compatibility assessment data, and rule adaptability assessment data are stored in a multi-dimensional feature format, constituting a set of descriptive data related to payment channel transmission. When extracting execution details of the clearing and settlement execution rules, the process reads the workflow connection data from historical interaction data to identify the collection, allocation, and transfer status of funds in the clearing and settlement process, obtaining fund processing status data; identifies the creation, modification, and review status of account records in the clearing and settlement process, obtaining account record status data; identifies the initiation, processing, and completion status of fund transfers in the clearing and settlement process, obtaining fund transfer status data; and identifies the generation, signing, and archiving status of vouchers in the clearing and settlement process, obtaining voucher generation status data. The above status data is stored in a structured format, constituting the detailed set of clearing and settlement execution rules.

[0020] Step S123: Construct the association mining network structure of the deep reinforcement learning co-evolution model. The association mining network structure includes feature input nodes, interaction simulation nodes, association extraction nodes, and association storage nodes.

[0021] In this embodiment, the association mining network structure is a hierarchical functional structure. The feature input node receives the set of data descriptions related to payment channel transmission and the set of details of clearing and settlement execution rules, maps the fields of the input data, and unifies fields with the same meaning from different sources into the same field name. After mapping, the data is transmitted to the interactive simulation node. The interactive simulation node receives the data transmitted by the feature input node, simulates the interaction process between payment channel transmission behavior and clearing and settlement execution rule processing behavior, and records the time nodes, state changes, and result feedback during the interaction process. The association extraction node receives the output data of the interactive simulation node and analyzes the association relationship between the data related to payment channel transmission and the clearing and settlement execution rules. The association storage node receives the output data of the association extraction node and stores the association data in the graph database. In the graph database, nodes represent data related to payment channel transmission and clearing and settlement execution rules, and edges represent the association relationship between the two. The attributes of the edges include the frequency of association occurrence and the corresponding interaction result.

[0022] Step S124: Align and fuse the payment channel transmission related data description set and the clearing and settlement execution rule detail set into the feature input nodes of the association mining network structure to obtain a fused dataset for simulation analysis.

[0023] In this embodiment, during alignment, the feature input node reads each piece of data from the payment channel transmission-related data description set and each piece of data from the clearing and settlement execution rule detail set, extracting the payment channel identifier, clearing and settlement execution rule identifier, and timestamp fields from each piece of data. Data with identical payment channel identifier, clearing and settlement execution rule identifier, and timestamp fields are matched. During fusion, the fields of the successfully matched payment channel transmission-related data description set and clearing and settlement execution rule detail set are concatenated, retaining all field content and not deleting any original data fields, to generate fused data. The fused data contains all description fields of the payment channel transmission-related data and all detail fields of the clearing and settlement execution rules. After generation, the fused data is sorted according to timestamp order to form a fused dataset, which is used for simulation analysis of subsequent interactive simulation nodes.

[0024] Step S125: Through the interactive simulation nodes of the association mining network structure, based on the fusion dataset, simulate the interaction process between the transmission behavior of the payment channel and the processing behavior of the clearing and settlement execution rules, and generate scenario-based interactive simulation results.

[0025] For example, step S1251: Extract core data describing the transmission characteristics of the payment channel and data describing the key links of the clearing and settlement execution rules from the fused dataset to obtain core transmission data and key link data.

[0026] In this embodiment, when extracting core transmission data, the payment channel transmission-related data description fields in the fused dataset are read, including transmission efficiency data, transmission stability data, process compatibility evaluation data, and rule adaptability evaluation data. These fields are integrated into core transmission data, which exists in the form of multi-dimensional feature vectors, with each vector dimension corresponding to a transmission characteristic indicator. When extracting key process data, the clearing and settlement execution rule detail fields in the fused dataset are read, including fund processing status data, account record status data, fund transfer status data, and voucher generation status data. These fields are integrated into key process data, which exists in structured data format, with each field corresponding to a key process status in the clearing and settlement process. After the core transmission data and key process data are extracted, they are synchronously stored in the input buffer of the interactive simulation node for subsequent simulation environment construction.

[0027] Step S1252: Based on the core transmission data and key link data, construct a scenario-based interactive simulation environment that includes a payment data transmission scenario simulation unit and a clearing and settlement process execution scenario simulation unit, and set up a data interaction channel between the payment data transmission scenario simulation unit and the clearing and settlement process execution scenario simulation unit.

[0028] For example, step S12521: Analyze the output data format and transmission protocol requirements of the payment data transmission scenario simulation unit to determine the basic specifications for data interaction.

[0029] In this embodiment, when analyzing the output data format, the output interface document of the payment data transmission scenario simulation unit is read, and the supported output data formats are extracted. These formats include field order, data type, and field length. When analyzing the transmission protocol requirements, the transmission protocol fields in the interface document are read, and the protocol type, port number, and transmission method are extracted. The output data format and transmission protocol requirements are integrated to determine the basic data interaction specifications. These specifications include data format version, protocol type, port number, field order, data type, and field length, and are stored in the specification database of the simulation environment construction module for subsequent compatibility verification.

[0030] Step S12522: Analyze the input data format and receiving protocol requirements of the settlement process execution scenario simulation unit, and compare them with the basic data interaction specifications to verify and achieve data interaction compatibility.

[0031] In this embodiment, when analyzing the input data format and receiving protocol requirements, the input interface document of the clearing and settlement process execution scenario simulation unit is read to extract the input data formats and receiving protocol requirements supported by the interface. During compatibility comparison, the input data format of the clearing and settlement process execution scenario simulation unit is compared with the output data format in the basic data interaction specification to verify whether the field order, data type, and field length are consistent. The receiving protocol requirements of the clearing and settlement process execution scenario simulation unit are compared with the transmission protocol requirements in the basic data interaction specification to verify whether the protocol type, port number, and transmission method are consistent. If the comparison is consistent, the compatibility verification passes; if the comparison is inconsistent, the input interface configuration of the clearing and settlement process execution scenario simulation unit is adjusted to be consistent with the basic data interaction specification to achieve data interaction compatibility.

[0032] Step S12523: Based on the basic data interaction specification information and the data reception compatibility specification information, construct the transmission architecture of the data interaction channel. This transmission architecture includes a data sending interface, a data receiving interface, and a data relay node.

[0033] In this embodiment, when constructing the transmission architecture, the network parameters of the data sending interface and the data receiving interface are configured according to the protocol type and port number in the basic data interaction specification. The data sending interface is deployed at the output end of the payment data transmission scenario simulation unit, and the data receiving interface is deployed at the input end of the clearing and settlement process execution scenario simulation unit. When configuring data relay nodes, the number and location of relay nodes are deployed according to the scale and frequency of the transmitted data. Relay nodes are used to store temporary transmitted data and provide data caching and forwarding services. The transmission architecture includes data sending interface identifier, data receiving interface identifier, relay node identifier, and network parameters, which are stored in the architecture database of the simulation environment construction module for subsequent interface configuration.

[0034] Step S12524: Set up a data format conversion unit at the data sending interface to convert the output data of the payment data transmission scenario simulation unit into a format compatible with the clearing and settlement process execution scenario simulation unit; set up a data verification unit at the data receiving interface to verify the format and integrity of the received data.

[0035] In this embodiment, when setting up a data format conversion unit, a format conversion module is deployed at the data sending interface. This module converts the output data of the payment data transmission scenario simulation unit into a compatible format based on the data interaction basic specifications and the input data format of the clearing and settlement process execution scenario simulation unit. The conversion rules include field mapping, data type conversion, and field length adjustment. When setting up a data verification unit, a verification module is deployed at the data receiving interface. This module performs format confirmation and integrity verification on the received data. Format confirmation verifies whether the field order, data type, and field length of the data meet the requirements. Integrity verification verifies whether the checksum of the data is correct. If the verification passes, the data is transmitted to the clearing and settlement process execution scenario simulation unit; if the verification fails, a retransmission mechanism is triggered.

[0036] Step S12525: Based on the functional positioning of the data relay node, deploy the data relay node to store temporary data during transmission and provide data caching and forwarding services.

[0037] In this embodiment, when deploying data relay nodes, corresponding relay servers are deployed in the simulated environment based on the relay node identifier and network parameters in the transmission architecture. Server configuration includes the number of CPU cores, memory size, and storage capacity. When configuring the relay node's functionality, caching and forwarding strategies are set. The caching strategy includes caching time and caching capacity, while the forwarding strategy includes forwarding order and forwarding frequency. After deployment, the relay node's caching and forwarding functions are tested. The tests include the read / write speed of cached data and the success rate of forwarded data. Once the tests are passed, the relay node is incorporated into the data interaction channel for temporary data storage and forwarding during transmission.

[0038] Step S12526: Based on the real-time transmission requirements of the data interaction channel, establish a real-time synchronization mechanism for data interaction, set the data transmission time interval and synchronization trigger conditions; configure the monitoring unit of the data interaction channel to monitor the data transmission rate, success rate and latency in real time, and record the transmission status.

[0039] In this embodiment, when establishing a real-time synchronization mechanism, the data transmission time interval is set according to the real-time transmission requirements of the data interaction channel. The time interval is determined based on the sampling frequency of the simulation operation. Synchronization trigger conditions are set, including data volume reaching a threshold and time interval arrival. Data synchronization is triggered when either condition is met. When configuring the monitoring unit, a monitoring server is deployed. The server connects to the data sending interface, data receiving interface, and data relay node to monitor the data transmission rate, success rate, and latency in real time. The monitoring data includes transmission rate values, success rate values, and latency values. The transmission status is recorded as normal, slow, or interrupted. The monitoring data and transmission status are stored in the monitoring database for subsequent anomaly handling.

[0040] Step S12527: Set up a data interaction anomaly handling process so that when the monitoring unit detects a transmission anomaly, it can automatically switch to a backup transmission path; perform stress testing on the data interaction channel to simulate high-concurrency data transmission scenarios to verify its transmission capacity and stability.

[0041] In this embodiment, when setting up the anomaly handling process, anomaly trigger thresholds are configured for the monitoring unit. These thresholds include a lower limit for transmission rate, a lower limit for success rate, and an upper limit for latency. When the monitored data exceeds these thresholds, a transmission anomaly is triggered. The anomaly handling process includes anomaly identification, backup path switching, and anomaly log recording. The backup path is a pre-configured redundant data interaction channel. During the switching process, the switching time, the cause of the anomaly, and the backup path identifier are recorded. When stress testing the data interaction channel, a high-concurrency data transmission scenario is simulated. The number and frequency of concurrently transmitted data packets are set to test the channel's transmission capacity and stability. Test metrics include maximum concurrency, transmission success rate, and latency. The test results are stored in the test database for channel performance evaluation and optimization.

[0042] Step S1253: Input the core transmission data into the payment data transmission scenario simulation unit and configure the basic operating parameters of the transmission scenario; input the key link data into the clearing and settlement process execution scenario simulation unit and configure the basic execution parameters of the clearing and settlement scenario.

[0043] In this embodiment, when inputting core transmission data, the input interface of the payment data transmission scenario simulation unit is called to convert the multi-dimensional feature vector of the core transmission data into a parameter format recognizable by the unit, and then input it into the unit's parameter configuration module. When configuring the basic operating parameters of the transmission scenario, the transmission delay range, response success rate range, and bandwidth limitation range are set as the unit's default operating parameters, and the parameter values ​​are determined based on the statistical characteristics of the core transmission data. When inputting key process data, the input interface of the clearing and settlement process execution scenario simulation unit is called to convert the structured fields of the key process data into a step configuration format recognizable by the unit, and then input it into the unit's step configuration module. When configuring the basic execution parameters of the clearing and settlement scenario, the step processing time range, state transition conditions, and anomaly trigger probability are set as the unit's default execution parameters, and the parameter values ​​are determined based on the historical execution records of the key process data.

[0044] Step S1254: Based on the basic operating parameters of the transmission scenario and the basic execution parameters of the clearing and settlement scenario, set up a data interaction channel between the payment data transmission scenario simulation unit and the clearing and settlement process execution scenario simulation unit; based on the data interaction channel and the basic operating parameters of the transmission scenario, start the scenario-based interactive simulation operation, the payment data transmission scenario simulation unit simulates the transmission process of payment data in the corresponding payment channel, and generates real-time transmission status data; based on the real-time transmission status data and the basic execution parameters of the clearing and settlement scenario, the clearing and settlement process execution scenario simulation unit receives the real-time transmission status data, simulates the corresponding clearing and settlement operation according to the requirements of the clearing and settlement execution rules, and generates real-time execution status data.

[0045] In this embodiment, when setting up the data interaction channel, the transmission protocol is configured as TCP, the data format as JSON, and the interaction frequency as once per second to ensure the reliability and real-time performance of data transmission. When starting the simulation operation, the execution command of the payment data transmission scenario simulation unit is first triggered. Based on the basic operating parameters of the transmission scenario, the unit simulates the transmission process of payment data from the initiating node to the receiving node, recording the node status, delay time, and data integrity during the transmission process, and generating real-time transmission status data. This real-time transmission status data includes a transmission stage identifier, the current node identifier, the delay time value, and the integrity identifier. The clearing and settlement process execution scenario simulation unit receives the real-time transmission status data through the data interaction channel. Based on the basic execution parameters of the clearing and settlement scenario and the requirements of the clearing and settlement execution rules, it simulates the corresponding clearing and settlement operations. These operations include triggering fund collection, updating account records, and preparing fund transfers. The unit records the step status, processing time, and result identifier during the operation process, generating real-time execution status data. This real-time execution status data includes the execution step identifier, the current status identifier, the processing time value, and the result identifier.

[0046] Step S1255: Based on real-time transmission status data and real-time execution status data, during the simulation operation, record the interaction trigger time, data interaction content and status changes between the payment data transmission scenario simulation unit and the settlement process execution scenario simulation unit, generate interaction process record data, and continuously run scenario-based interactive simulation operations based on the interaction process record data to cover multiple simulation scenarios with different payment scales and different transmission timeliness requirements, and generate interaction process record data under multiple scenarios.

[0047] In this embodiment, when recording the interaction process, a timestamp sampling interval is set. At each sampling interval, the interaction trigger time, data interaction content, and state changes of the two simulation units are recorded once. The interaction trigger time is the transmission timestamp of the data interaction channel; the data interaction content includes the real-time transmission status data and real-time execution status data; and the state changes are the changes in the running state of the two simulation units before and after the data interaction. The above recorded content is integrated into interaction process recording data, which includes a timestamp, trigger time, interaction content summary, and state change summary. During continuous simulation operation, the payment scale parameters and transmission timeliness requirement parameters of the payment data transmission scenario simulation unit are adjusted. The parameter adjustment range covers the minimum to maximum values ​​in historical interaction data. After each adjustment, the simulation operation is restarted, and the corresponding interaction process recording data is recorded, generating a multi-scenario interaction process recording data set. This set is stored in the simulation result database for subsequent aggregation of scenario-based interaction simulation results.

[0048] Step S1256: Summarize the interaction process recording data under the multiple scenarios to generate the scenario-based interaction simulation results.

[0049] In this embodiment, when summarizing the interaction process record data, the data set of interaction process record data from multiple scenarios is read, and the data is grouped according to the simulation scenario identifier, with each group corresponding to a simulation scenario. Statistical analysis is performed on the data within each group to calculate the interaction trigger frequency, data interaction success rate, and number of state changes. The statistical results serve as feature indicators for the scenario-based interaction simulation results. The grouped data and statistical results are integrated to generate scenario-based interaction simulation results. The results include simulation scenario identifiers, a list of interaction process record data, and statistical values ​​of feature indicators. The results are stored in the output buffer of the interaction simulation node for subsequent extraction of associated data from related extraction nodes.

[0050] Step S126: Using the association extraction nodes of the association mining network structure, explicit and implicit association data are separated and extracted from the scenario-based interactive simulation results to obtain the separated and extracted association data. The implicit association data in the separated and extracted association data is structured and stored through the association storage nodes of the association mining network structure to generate a set of implicit association data with payment channel transmission related data as the core index, clearing and settlement execution rules as the association clues, and interactive effect data as the association strength identifier.

[0051] In this embodiment, when extracting associated data, the associated extraction node reads the scenario-based interactive simulation results. The correspondence between payment channel transmission data and clearing / settlement execution rules that directly appear in the interaction timeline is taken as explicit associated data. The causal relationship between changes in payment channel transmission status and changes in clearing / settlement processing status in the interaction timeline is analyzed, and correspondences that do not directly appear but are derived through causal inference are taken as implicit associated data. The extracted associated data includes both explicit and implicit associated data. During the structured storage processing of implicit associated data, the identifier of the payment channel transmission-related data is used as the core index, and the corresponding clearing / settlement execution rule identifier is associated. The interaction result identifier in the interactive simulation results is used as the association strength identifier to generate implicit associated data records. Implicit associated data records include a core index field, an association clue field, and a strength identifier field. Multiple records constitute an implicit associated data set, which is stored in a dedicated node and edge structure of the graph database. The core index corresponds to a graph node, the association clue corresponds to the target node of the edge, and the strength identifier corresponds to the attribute value of the edge.

[0052] Step S127: Establish a data connection between the implicit association data set and the decision output node of the deep reinforcement learning co-evolution model, and set up a data feedback receiving channel at the evolution update node of the deep reinforcement learning co-evolution model. The data feedback receiving channel is used to receive the transmission data of the payment routing dynamic adjustment path and the execution data of the clearing and settlement process adaptation strategy collected subsequently.

[0053] In this embodiment, when establishing a data connection, the input interface of the decision output node is configured to read the storage address and data format of the implicitly related data set. Records in the implicitly related data set are read through the interface call for reference in the decision output process. When setting up the data feedback receiving channel, a data receiving service is deployed on the evolution update node, and the service's listening port and data format parsing rules are configured. The data format parsing rules include field order, data type, and checksum calculation method to ensure that the received transmitted and executed data can be correctly parsed. The configuration information of the data feedback receiving channel includes the channel identifier, listening address, port number, and data format version. After configuration, the channel identifier is synchronized to the data acquisition module for subsequent targeted transmission of feedback data.

[0054] Step S130: Analyze the interaction between payment channel transmission data and clearing and settlement execution rules through a deep reinforcement learning collaborative evolution model, and generate dynamic adjustment paths for payment routing and adaptation strategies for clearing and settlement processes.

[0055] In this embodiment, when analyzing the interaction relationship, the payment channel transmission data and clearing and settlement execution rule data are first input into the information perception node of the deep reinforcement learning co-evolution model. After processing, the information perception node transmits the data to the interaction simulation node, which simulates the interaction process between the two and generates the interaction simulation result. The decision output node receives the interaction simulation result and, combined with the correlation relationship in the implicit correlation data set, analyzes the interaction effect of different combinations of payment channels and clearing and settlement execution rules. Based on the effect evaluation result, it generates a dynamic adjustment path for payment routing and an adaptation strategy for the clearing and settlement process. The dynamic adjustment path for payment routing includes the transmission node sequence, node switching conditions, and path priority. The adaptation strategy for the clearing and settlement process includes the execution step order, step triggering conditions, and exception handling rules. After generation, the two are associated and bound, and the binding relationship is stored in the strategy database for subsequent linkage invocation during the execution process.

[0056] Step S131: Parse the key information contained in the received payment request and generate an adaptation target description for payment routing and clearing and settlement. The key information includes data on the initiation scenario of the payment request, data on the scale of fund transfer, data on transmission timeliness requirements, and data on related business types.

[0057] In this embodiment, when parsing key information of a payment request, the initiation scenario field in the payment request data packet is read to obtain initiation scenario data; the fund transfer scale field is read to obtain fund transfer scale data; the transmission timeliness requirement field is read to obtain transmission timeliness requirement data; and the associated business type field is read to obtain associated business type data. When generating the adaptation target representation, the above key information data is converted into a combination of natural language and structured instructions. The natural language part is used for the readability of the strategy, and the structured instructions part is used for machine reading of the model. The adaptation target representation includes a target identifier, key information structured data, and a natural language description. After generation, it is stored in the target cache for reference in the generation of subsequent adaptation strategies.

[0058] Step S132: Based on the adaptation target description of payment routing and clearing and settlement, retrieve the implicit correlation data set stored in the deep reinforcement learning co-evolution model, filter out payment channel transmission related data entries and clearing and settlement execution rule data entries that have a high correlation with the key information contained in the received payment request, obtain a set of matched data entries, input the payment channel transmission related data and clearing and settlement execution rule data in the set of matched data entries into the information perception node of the deep reinforcement learning co-evolution model for real-time data perception and feature extraction processing, and obtain the corresponding feature data.

[0059] In this embodiment, when retrieving the implicitly related data set, the key information structured data in the target description is used as the search condition. The core index fields and related clue fields in the implicitly related data set are matched, and the matching degree of each entry is calculated. The matching degree calculation is based on the overlap rate and weight coefficient of the field content; entries with a high overlap rate and a large weight coefficient have a high matching degree. When filtering highly related entries, a matching degree threshold is set, and entries with a matching degree higher than the threshold are included in the set of matched data entries. During feature extraction, the information perception node reads the set of matched data entries, extracts the numerical fields and categorical fields from each entry, standardizes the numerical fields, and encodes the categorical fields. The processed data serves as feature data, containing numerical feature vectors and categorical feature codes, which are used as input for subsequent interactive simulation nodes.

[0060] Step S133: The information sensing node transmits the feature data to the interactive simulation node of the deep reinforcement learning co-evolution model. The interactive simulation node simulates the interaction process under different combinations of payment channels and different clearing and settlement processes, and evaluates and obtains the interaction effect evaluation data of each combination.

[0061] In this embodiment, during the simulated interaction process, the interactive simulation node receives feature data and generates different interactive scenarios based on the payment channel identifier and clearing and settlement execution rule identifier in the feature data. Each scenario corresponds to a combination of payment channels and clearing and settlement processes. When simulating each scenario, the transmission process of the payment request and the clearing and settlement process are simulated in chronological order, and the state changes and results of each process are recorded. When evaluating the interaction effect, based on the recorded state changes and results, the transmission efficiency score, execution smoothness score, and resource consumption score for each combination are calculated. The score calculation is based on a preset scoring standard, which includes indicator weights and score ranges. The weighted sum of the three scores serves as the interaction effect evaluation data. The evaluation data includes the combination identifier, transmission efficiency score, execution smoothness score, resource consumption score, and comprehensive score.

[0062] Step S134: Based on the combined interaction effect data, the decision output node of the deep reinforcement learning co-evolution model initially generates multiple payment routing candidate paths and corresponding clearing and settlement candidate processes, thus obtaining a set of initially generated candidate paths and candidate processes.

[0063] For example, step S1341: parse the combined interaction effect evaluation data and extract the transmission efficiency index, execution smoothness index and resource consumption index corresponding to each combination.

[0064] In this embodiment, when parsing the combined interaction effect evaluation data, the evaluation data set output by the interaction simulation node is read. Each evaluation data entry corresponds to a combination of a payment channel and a clearing and settlement process. When extracting the transmission efficiency index, the transmission time statistics and response success rate statistics from the evaluation data entries are read, and the product of the reciprocal of the transmission time statistics and the response success rate statistics is used as the transmission efficiency index. When extracting the execution smoothness index, the step processing time statistics and state transition power statistics from the evaluation data entries are read, and the product of the reciprocal of the step processing time statistics and the state transition power statistics is used as the execution smoothness index. When extracting the resource consumption index, the bandwidth usage statistics and CPU usage statistics from the evaluation data entries are read, and the weighted sum of the bandwidth usage statistics and CPU usage statistics is used as the resource consumption index. After the above indexes are extracted, they are stored in the index cache area of ​​the decision output node for subsequent comprehensive evaluation.

[0065] Step S1342: Based on the transmission efficiency index, execution smoothness index and resource consumption index of each combination, a comprehensive evaluation is performed on each combination. Based on the comprehensive evaluation results, combinations that meet the preset conditions are selected to generate a candidate combination set.

[0066] In this embodiment, when comprehensively evaluating the combinations, weight coefficients are assigned to the transmission efficiency index, execution smoothness index, and resource consumption index. These weight coefficients are determined based on feedback data of historical decision-making effects. A comprehensive evaluation score is calculated for each combination by summing the scores of the transmission efficiency index multiplied by its weight coefficient, the execution smoothness index multiplied by its weight coefficient, and the resource consumption index multiplied by its weight coefficient. A preset threshold for the comprehensive evaluation score is set, and combinations with scores higher than the threshold are selected to generate a candidate combination set. This set contains a combination identifier, a comprehensive evaluation score, and scores for each index, and is stored in the candidate set cache of the decision output node for subsequent generation of candidate paths and processes.

[0067] Step S1343: For each combination in the candidate combination set, extract its corresponding payment channel configuration information and clearing and settlement rule information. For each combination, based on its payment channel configuration information, arrange the transmission node sequence of the payment request from the initiator to the receiver to generate a preliminary payment route candidate path. For each combination, based on its clearing and settlement rule information, arrange the steps of fund processing, account recording, fund transfer and voucher generation to generate a corresponding clearing and settlement candidate process.

[0068] In this embodiment, when extracting configuration and rule information, the identifier of each combination in the candidate combination set is read, and the payment channel configuration database and clearing and settlement rule database are queried according to the identifier to obtain the corresponding payment channel configuration information and clearing and settlement rule information. The payment channel configuration information includes a node list, inter-node transmission protocol, and node bandwidth limits; the clearing and settlement rule information includes a step list, step triggering conditions, and step processing requirements. When generating a preliminary payment route candidate path, the transmission node sequence is arranged according to the physical connection order from the initiator to the receiver based on the node list in the payment channel configuration information. The sequence includes node identifier, inter-node transmission protocol, and transmission priority to generate a preliminary payment route candidate path. When generating a clearing and settlement candidate process, the step sequence is arranged according to the logical order of fund processing, account recording, fund transfer, and voucher generation based on the step list in the clearing and settlement rule information. The sequence includes step identifier, step triggering conditions, and step processing requirements to generate the corresponding clearing and settlement candidate process.

[0069] Step S1344: Associate the preliminary payment route candidate path generated by each combination with its corresponding clearing and settlement candidate process, determine the triggering relationship and connection node between the path and the process, and complete the associated path and process, and improve the missing node information and step information.

[0070] In this embodiment, when associating paths and processes, the terminal node identifiers of the preliminary payment routing candidate paths are matched with the triggering conditions of the starting steps of the clearing and settlement candidate processes to determine the triggering relationship of the process starting steps after the path transmission is completed. The processing results of each step of the clearing and settlement candidate processes are matched with the node states of the preliminary payment routing candidate paths to determine the impact of process step processing on the path node states. When determining connecting nodes, key nodes in the path are associated with key steps in the process. Key nodes are data forwarding nodes in the path, and key steps are fund transfer steps in the process. The association includes node triggering steps and step feedback nodes. During completeness supplementation, it is checked whether the node information in the preliminary payment routing candidate paths contains all necessary fields. Necessary fields include node identifier, transmission protocol, and bandwidth limit; if missing, default values ​​are added. It is also checked whether the step information in the clearing and settlement candidate processes contains all necessary fields. Necessary fields include step identifier, triggering conditions, and processing requirements; if missing, default values ​​are added.

[0071] Step S1345: For each combination in the candidate combination set, perform the above-mentioned path generation, process generation, association and supplementation operations to generate multiple payment routing candidate paths and corresponding clearing and settlement candidate processes. Sort the generated multiple payment routing candidate paths and clearing and settlement candidate processes according to the comprehensive evaluation results of their corresponding combinations to generate the initially generated candidate path and candidate process set.

[0072] In this embodiment, when performing the path and process generation operation, steps S1343 to S1344 are repeated for each combination in the candidate combination set to generate corresponding preliminary payment route candidate paths and clearing and settlement candidate processes. When sorting the candidate paths and processes, the comprehensive evaluation score of each combination is read, and the generated candidate paths and processes are sorted in descending order of score. The sorting result includes combination identifier, candidate path identifier, candidate process identifier, and comprehensive evaluation score, generating a preliminary set of candidate paths and candidate processes. The set is stored in the candidate set database of the decision output node for subsequent adaptation analysis.

[0073] Step S135: Based on the combined interaction effect evaluation data and the correlation strength identifier in the implicit correlation data set, perform a fit analysis on the multiple payment route candidate paths and clearing and settlement candidate process combinations in the initially generated candidate path and candidate process set, and screen out candidate combinations whose fit meets the preset threshold to obtain the screened candidate combinations.

[0074] In this embodiment, during the fit analysis, the comprehensive score from the combined interaction effect evaluation data and the correlation strength identifier from the implicit correlation data set are read. The comprehensive score and the strength identifier are weighted and calculated to obtain the fit value. The weight coefficients for the weighted calculation are determined based on the model training results. When screening candidate combinations, a fit threshold is set. Combinations with fit values ​​higher than the threshold are selected as candidate combinations. The selected candidate combinations include a combination identifier, a payment routing candidate path, a clearing and settlement candidate process, and a fit value. The set is arranged in descending order of fit value and is used for the generation of subsequent dynamic adjustment logic.

[0075] Step S136: For the selected candidate combinations, the dynamic adjustment logic is generated through the decision output node of the deep reinforcement learning collaborative evolution model to determine the adjustment trigger conditions and adjustment methods of payment routing under different transmission scenarios, thus obtaining the dynamic adjustment logic.

[0076] For example, step S1361: parse the filtered candidate combinations, extract the path features of the payment routing candidate paths and the process features of the clearing and settlement candidate processes contained in each candidate combination. The path features include the transmission node sequence and node connection relationship, and the process features include the execution step sequence and step connection relationship.

[0077] In this embodiment, when parsing candidate combinations, each combination entry in the filtered candidate combination set is read, and the corresponding payment routing candidate path identifier and clearing and settlement candidate process identifier are extracted. When extracting path features, the path database is queried based on the payment routing candidate path identifier to obtain the transmission node sequence and node connection relationships. The transmission node sequence contains the sequential arrangement of node identifiers, and the node connection relationships contain the transmission protocols and bandwidth limitations between nodes. When extracting process features, the process database is queried based on the clearing and settlement candidate process identifier to obtain the execution step sequence and step connection relationships. The execution step sequence contains the sequential arrangement of step identifiers, and the step connection relationships contain the triggering conditions and feedback mechanisms between steps. The path features and process features are integrated and stored in the feature database of the decision output node for subsequent scenario simulation.

[0078] Step S1362: Based on the path features and process features, simulate various transmission scenario change factors including fluctuations in payment channel availability, changes in channel traffic load, and fluctuations in payment request timeliness requirements, generate a simulated transmission scenario dataset, import the simulated transmission scenario dataset into the interactive simulation node of the deep reinforcement learning co-evolution model, calculate the state change process of each candidate combination under the influence of different change factors, and identify potential path interruption risks and process blocking risks.

[0079] In this embodiment, when simulating factors affecting transmission scenarios, the availability of payment channels is simulated based on the transmission node sequence and node connection relationships in the path characteristics. The fluctuation range includes changes in node availability. Channel traffic load changes are also simulated, with the range including changes in node bandwidth utilization. Payment request timeliness requirements are simulated, with the range including changes in request timeout time. These simulated factors are integrated to generate a simulated transmission scenario dataset, which includes scenario identifiers, change factor types, and change ranges. During the state change process, the simulated transmission scenario dataset is imported into the interactive simulation node. The node simulates the transmission and execution state changes of each candidate combination under the influence of different factors, recording the time points, magnitudes, and impacts of the state changes. When identifying potential risks, the transmission interruption time and process blocking time during the state change process are analyzed. Interruption times exceeding a threshold are identified as path interruption risks, and blocking times exceeding a threshold are identified as process blocking risks. The risk identification results include combination identifiers, scenario identifiers, risk types, and risk node / step identifiers, and are stored in the risk database of the decision output node for subsequent mapping relationship generation.

[0080] Step S1363: Associate the potential path interruption risk and process blocking risk with specific transmission nodes in the path features, and associate the potential process blocking risk with specific execution steps in the process features to generate a mapping relationship between risk nodes and risk steps. Based on the mapping relationship between risk nodes and risk steps, retrieve relevant historical adjustment records involving similar transmission scenarios and clearing and settlement rules from the historical interaction patterns stored in the interactive simulation nodes. The historical adjustment records contain descriptions of historical trigger scenarios and descriptions of historical adjustment actions to obtain relevant historical adjustment records.

[0081] In this embodiment, when generating the mapping relationship, the risk node / step identifier in the risk identification result is read, the path interruption risk is associated with the specific transmission node in the path features, and the process blockage risk is associated with the specific execution step in the process features, generating a mapping relationship between risk nodes and risk steps. The relationship includes risk identifier, node / step identifier, and risk type. When retrieving historical adjustment records, based on the risk type and node / step identifier in the mapping relationship, historical adjustment records of similar transmission scenarios and clearing and settlement rules are retrieved from the historical interaction mode database of the interactive simulation node. Similarity matching is based on the overlap of risk type, node / step identifier, and scenario features. Records with an overlap higher than a threshold are considered relevant historical adjustment records. These records include a description of the historical trigger scenario, a description of the historical adjustment action, and an evaluation of the adjustment effect, and are stored in the historical record database of the decision output node for subsequent rule framework construction.

[0082] Step S1364: Based on the relevant historical adjustment records, construct a rule structure framework that includes condition units and execution units. The condition units are used to describe the adjustment trigger conditions, and the execution units are used to describe the adjustment methods.

[0083] In this embodiment, when constructing the rule structure framework, the historical trigger scenario descriptions and historical adjustment action descriptions in the relevant historical adjustment records are analyzed. Key conditions in the historical trigger scenario descriptions are extracted as fields for condition units, including scenario type, node / step identifier, and threshold range. Key actions in the historical adjustment action descriptions are extracted as fields for execution units, including adjustment type, adjustment object, and adjustment parameters. The condition units and execution units are integrated to construct the rule structure framework. The framework includes a list of condition unit fields, a list of execution unit fields, and the association method between conditions and execution. This framework is stored in the rule framework database of the decision output node for subsequent condition filling.

[0084] Step S1365: For each risk node and risk step in the mapping relationship between risk nodes and risk steps, fill in the specific attributes of the condition unit based on the rule structure framework. The specific attributes include risk node identifier, risk step identifier, and threshold range of associated change factors to constitute adjustment trigger conditions.

[0085] In this embodiment, when filling in the condition unit attributes, the risk node identifier, risk step identifier, and risk type are read from the mapping relationship between risk nodes and risk steps. According to the condition unit field list in the rule structure framework, the risk node identifier and risk step identifier are filled into the corresponding fields. The threshold range is determined according to the fluctuation range of the associated change factors and filled into the corresponding fields to form the adjustment trigger condition. The condition includes the risk node identifier, risk step identifier, threshold range, and risk type, and is stored in the trigger condition database of the decision output node for subsequent action matching.

[0086] Step S1366: For each adjustment trigger condition, match the historical adjustment actions in the relevant historical adjustment records, and extract the path modification actions and process modification actions. The path modification actions include node replacement and sequence rearrangement, and the process modification actions include step skipping and step addition, to constitute the adjustment method.

[0087] In this embodiment, when matching historical adjustment actions, the risk type, risk node identifier, and risk step identifier in the adjustment trigger conditions are read. The corresponding historical adjustment action description is retrieved from relevant historical adjustment records, based on the overlap of risk type and node / step identifier. When extracting adjustment actions, path modification actions and process modification actions are extracted from the historical adjustment action descriptions. Path modification actions include the target node identifier for node replacement and the new node order for sequence rearrangement. Process modification actions include the identifier of skipped steps and the content added to the steps. The extracted adjustment actions are integrated to form an adjustment method, which includes a list of path modification actions and a list of process modification actions. These are stored in the adjustment method database of the decision output node for subsequent rule chain generation.

[0088] Step S1367: Associate and bind each adjustment trigger condition with its corresponding adjustment method to generate an adjustment rule chain. For each candidate combination in the filtered candidate combinations, repeat the above-mentioned feature extraction, scenario simulation, risk identification, record retrieval, framework construction, condition filling, action matching and rule chain generation operations to generate an adjustment rule chain set corresponding to each candidate combination.

[0089] In this embodiment, when associating trigger conditions with adjustment methods, the identifier of the adjustment trigger condition is bound to the identifier of the corresponding adjustment method to generate an adjustment rule chain. The rule chain includes a condition identifier, a method identifier, and an association priority. When repeating the operation, for each of the filtered candidate combinations, steps S1361 to S1366 are repeated to generate a corresponding adjustment rule chain. The adjustment rule chains of all combinations are integrated to generate an adjustment rule chain set corresponding to each candidate combination. The set includes a combination identifier, a rule chain list, and a rule chain priority, and is stored in the rule chain database of the decision output node for subsequent integration.

[0090] Step S1368: Based on the rule chain set corresponding to multiple candidate combinations, integrate the rule chain set corresponding to multiple candidate combinations into the rule base of the decision output node of the deep reinforcement learning co-evolution model to complete the generation of dynamic adjustment logic.

[0091] In this embodiment, when integrating the rule chain set, the adjustment rule chain set corresponding to each candidate combination is read, and the rule chains are sorted according to their priority. Rule chains with higher priority are stored in the rule base first. When configuring the rule base of the decision output node, the sorted rule chains are imported into the rule base, and the retrieval method and matching mechanism of the rule base are set. The retrieval method includes conditional matching and keyword matching, and the matching mechanism includes exact matching and fuzzy matching. After integration, the rule invocation function of the rule base is tested. The test content includes rule retrieval speed and matching accuracy. After the test passes, the dynamic adjustment logic is generated and stored in the dynamic adjustment logic database of the decision output node for subsequent path refinement.

[0092] Step S137: Based on the adjustment triggering conditions and adjustment methods in the dynamic adjustment logic, refine the node connection details and transmission order of the payment route candidate paths in the filtered candidate combinations, and generate the payment route dynamic adjustment path.

[0093] In this embodiment, when refining the candidate payment routing path, the adjustment trigger conditions and adjustment methods in the dynamic adjustment logic are read. For each node in the candidate path, node replacement rules and path switching rules corresponding to the trigger conditions are added; for the transmission order between nodes, order adjustment rules under peak traffic scenarios are added. The refined path includes node sequence, node connection rules, transmission order rules, adjustment trigger conditions, and adjustment methods. After generation, the path validity is verified, including node reachability and transmission protocol compatibility. Once the verification is passed, it is used as the dynamically adjusted payment routing path and stored in the routing database for subsequent payment request routing guidance.

[0094] Step S138: Based on the transmission characteristics of the payment routing dynamic adjustment path, synchronously match the execution steps and connection nodes of the clearing and settlement candidate processes in the filtered candidate combinations, generate a clearing and settlement process adaptation strategy that is compatible with the payment routing dynamic adjustment path, associate and bind the payment routing dynamic adjustment path with the clearing and settlement process adaptation strategy, and generate a complete output result containing the path and strategy correspondence.

[0095] In this embodiment, when matching candidate clearing and settlement processes, the transmission characteristics of the dynamically adjusted payment routing path are analyzed. These transmission characteristics include transmission delay range, node switching frequency, and data integrity requirements. The timeout threshold for each step in the clearing and settlement process is adjusted based on the transmission delay range; the fault tolerance mechanism for connecting nodes is adjusted based on the node switching frequency; and the voucher generation rules are adjusted based on the data integrity requirements. The matched clearing and settlement process includes the execution step sequence, step timeout threshold, fault tolerance mechanism for connecting nodes, and voucher generation rules, which are then used as the clearing and settlement process adaptation strategy. When associating and binding paths and strategies, the identifier of the dynamically adjusted payment routing path is associated with the identifier of the clearing and settlement process adaptation strategy, generating a corresponding relationship record. This record includes the path identifier, strategy identifier, transmission characteristic summary, and process adaptation summary. Multiple records constitute the complete output result, which is stored in the strategy output database for subsequent execution process calls.

[0096] Step S140: Based on the payment routing, dynamically adjust the path to guide the payment request to complete the data transmission within the corresponding payment channel, and simultaneously execute the entire clearing and settlement process operation according to the clearing and settlement process adaptation strategy.

[0097] In this embodiment, during data transmission, the node sequence and connection rules of the payment routing dynamic adjustment path are read, and the routing field of the payment request data packet is set as the identifier of the first node of the path, and sent to the node of the corresponding payment channel. After receiving the data packet, the node determines whether the adjustment condition is triggered according to the connection rules. If not, it forwards the packet to the next node; if triggered, it performs node replacement or path switching according to the adjustment method until the data packet is transmitted to the target node. During the synchronous execution of the clearing and settlement process, the execution step sequence and triggering conditions of the clearing and settlement process adaptation strategy are read. When the payment request data packet is transmitted to the target node, the first step of the clearing and settlement process is triggered, and the steps are executed sequentially according to the step sequence. During the execution, the start time, end time, and result status of each step are recorded until all steps are completed.

[0098] Step S141: Extract the payment channel identifier data and path node data contained in the payment routing dynamic adjustment path, determine the target payment channel and transmission node order that the payment request needs to access, and obtain the target payment channel information and transmission node order information.

[0099] In this embodiment, when extracting identifier and node data, the stored record of the dynamically adjusted payment route is read. The content of the payment channel identifier field in the record is used as the payment channel identifier data, and the content of the node sequence field is used as the path node data. When determining the target payment channel, the payment channel identifier data is matched with the channel list in the payment system to obtain the name, interface address, and transmission protocol information of the target payment channel, thus constituting the target payment channel information. When determining the transmission node order, the path node data is arranged in order to obtain the transmission node order information. The order information includes the node identifier, node interface address, and inter-node transmission protocol, which is used for the establishment of subsequent transmission connections and the setting of routing guidance.

[0100] Step S142: Based on the target payment channel identifier data in the target payment channel information, establish a transmission connection between the payment request and the target payment channel, and set the data transmission routing guidance information according to the transmission node sequence information to obtain the transmission connection and routing guidance information.

[0101] In this embodiment, when establishing a transmission connection, the channel configuration information of the payment system is queried based on the target payment channel identifier data to obtain the access interface address and authentication parameters of the channel. A connection request is sent to the interface address using the authentication parameters. After the interface returns a connection success identifier, a persistent transmission connection is established. When setting routing guidance information, the transmission node sequence information is converted into routing instructions. The routing instructions include the node sequence identifier, the forwarding address of each node, and the forwarding timeout. The routing instructions are added to the header of the payment request data packet to constitute the routing guidance information. The transmission connection and routing guidance information include a connection identifier, an interface address, and routing instructions. After the information is generated, it is stored in the connection management module for subsequent data packet sending and forwarding.

[0102] Step S143: Integrate the key information of the payment request with the routing guidance information to generate a data packet containing the transmission instructions, and send the data packet containing the transmission instructions to the first transmission node of the target payment channel through the established transmission connection.

[0103] In this embodiment, when integrating information, the key information field of the payment request and the routing instruction field of the routing guidance information are read. The routing instruction field is added to the header of the payment request data packet, while the key information field is retained in the payload area of ​​the data packet, generating a data packet containing transmission instructions. When sending the data packet, the sending interface of the transmission connection is called to send the data packet to the first transmission node of the target payment channel. During the sending process, the sending time, data packet size, and connection status are recorded. After the sending is completed, the node's reception confirmation information is awaited. The confirmation information includes the reception time and data packet integrity identifier, which is used for the generation of subsequent transmission records.

[0104] Step S144: Each transmission node of the target payment channel receives and forwards data packets containing transmission instructions in sequence according to the routing guidance information. During the forwarding process, the transmission time data and data integrity data of each node are recorded to obtain the transmission record data of each node.

[0105] In this embodiment, when nodes receive and forward data packets, the first transmitting node receives the data packet and verifies its integrity. The verification includes checksum calculation and field integrity checks. Upon successful verification, the receiving time and integrity flag are recorded as transmission time data and data integrity data, respectively. Then, based on the next node address in the routing guidance information, the data packet is forwarded to the next transmitting node. During forwarding, the forwarding time and forwarding status are recorded. Subsequent nodes repeat the above receiving, verification, recording, and forwarding operations until the data packet is transmitted to the terminal node. The transmission record data of each node includes the node identifier, receiving time, forwarding time, integrity flag, and forwarding status. Multiple records constitute the transmission record data set of each node, which is stored in the transmission log database for subsequent collection and analysis of transmission data.

[0106] Step S145: When the data packet containing the transmission instruction is transmitted to the terminal node of the target payment channel, the execution instruction of the clearing and settlement process adaptation strategy is triggered. Based on the execution instruction of the clearing and settlement process adaptation strategy, and in accordance with the execution steps included in the clearing and settlement process adaptation strategy, the fund collection operation is first performed to collect the fund data corresponding to the payment request and classify and organize it to obtain the classified and organized fund data. Based on the classified and organized fund data, the account recording operation is performed to record the fund flow details and account balance data in accordance with the requirements of the clearing and settlement execution rules to obtain the fund flow details data and account balance data.

[0107] For example, step S1451: Analyze the specific requirements for fund collection operations in the settlement process adaptation strategy, determine the scope, method and data recording requirements of fund collection, and obtain the fund collection operation requirement information.

[0108] In this embodiment, when parsing the fund collection operation requirements, the fund collection step entries of the clearing and settlement process adaptation strategy are read, and the scope field, method field, and data recording requirement field are extracted from the entries. The scope field includes the type of fund source account, the region of the account, and the time range of fund deposit; the method field includes the collection triggering method, collection frequency, and collection amount limit; the data recording requirement field includes the record field list, record format requirements, and record storage location. The above field contents are integrated to generate fund collection operation requirement information, which includes scope information, method information, and data recording requirement information. This information is stored in the operation requirement cache area of ​​the clearing and settlement system for subsequent fund collection operations.

[0109] Step S1452: Based on the fund collection operation requirements and key information of the payment request, locate the fund source account information and fund amount information to be collected, obtain the fund source and amount information, and extract the corresponding amount of fund information from each fund source account through the fund query interface of the payment system. The extracted information includes the fund deposit time, fund status and account identifier, and obtain the original information of the extracted funds.

[0110] In this embodiment, when locating the source and amount of funds, the scope information and key information of the payment request in the fund collection operation requirements are read. The fund source account type, account location, and fund deposit time range are matched to determine the fund source account identifier that needs to be collected. Based on the payment amount information in the key information of the payment request, the amount of funds to be collected for each fund source account is determined. When extracting fund information through the fund query interface, the fund query interface of the payment system is called. The fund source account identifier and the amount of funds to be collected are input. The interface returns the fund deposit time, fund status, and account identifier of the corresponding account. The returned information is integrated into the extracted raw fund information, which includes the account identifier, fund amount, deposit time, and fund status. This information is stored in the raw fund information database of the clearing and settlement system for subsequent classification and organization.

[0111] Step S1453: Classify the extracted original fund information according to the type of fund source account, distinguish the fund information corresponding to different business types and different initiators, and obtain classified fund information. Based on the classified fund information and the key information of the payment request, check the classified fund information one by one, record the key information of each fund, and verify whether the fund information is consistent with the key information of the payment request during the verification process. The fund information that is consistent with the verification is taken as the verified fund information.

[0112] In this embodiment, when classifying the original fund information, the account identifier in the extracted original fund information is read. Based on the account type corresponding to the account identifier, the original fund information is divided into personal account fund information, corporate account fund information, and third-party account fund information. Fund information for different business types and initiators is stored in the corresponding classification sub-databases. When verifying the fund information, the fund amount, deposit time, and account identifier in the classified fund information are read and compared with the payment amount, initiation time, and initiating account identifier in the key information of the payment request, and the two are compared one by one to verify whether they are consistent. The verification criteria for consistency are that the fund amount is equal, the deposit time is within the preset range of the initiation time, and the account identifier matches. Fund information that meets the criteria is used as the verified fund information and stored in the verified fund information database of the clearing and settlement system for subsequent processing.

[0113] Step S1454: Based on the verified fund information and the fund collection operation requirements, organize the verified fund information according to the format required by the clearing and settlement process adaptation strategy to generate a structured fund collection data table. Add the execution time and operation node information of the collection operation to the structured fund collection data table to improve the data record content.

[0114] In this embodiment, when organizing fund information, the data record requirement fields in the fund collection operation requirement information are read. Following the required field list and format, the account identifier, fund amount, deposit time, and fund status from the verified fund information are filled into the corresponding fields to generate a structured fund collection data table. When completing the data record content, the current system time is read as the execution time of the collection operation, and the node identifier of the fund collection operation is read as the operation node information. This information is added to the corresponding fields of the fund collection data table. The data table contains account identifier, fund amount, deposit time, fund status, execution time, and operation node information, and is stored in the fund collection data table database of the clearing and settlement system for subsequent completion notifications of fund collection operations.

[0115] Step S1455: Store the improved fund collection data table in the dedicated database of the clearing and settlement system, and set the corresponding data access permissions and data backup mechanism. Based on the improved fund collection data table, generate a fund collection operation completion notification. The fund collection operation completion notification includes the total amount of collected funds, the number of collected accounts, and the collection completion time information.

[0116] In this embodiment, when storing the funds collection data table, the improved data table is written into a dedicated database of the clearing and settlement system. This dedicated database uses encrypted storage, and the storage key is changed periodically. When setting data access permissions, user roles and permission levels are configured for the database, ensuring that only the core operation nodes and audit nodes of the clearing and settlement system have access. When setting a data backup mechanism, the automatic backup frequency and backup storage location are configured, and backup data is periodically synchronized to an off-site backup server. When generating a funds collection operation completion notification, the fund amount field in the improved funds collection data table is read, and the sum of all fields is calculated as the total collected funds. The account identifier field in the data table is read, and the number of different account identifiers is counted as the number of collected accounts. The execution time field in the data table is read, and the latest execution time is used as the collection completion time. The completion notification includes the total collected funds, the number of collected accounts, and the collection completion time, and is stored in the clearing and settlement system's operation completion notification database for triggering subsequent account recording operations.

[0117] Step S1456: Transmit the completion notification of the fund collection operation to the next execution step of the clearing and settlement process to trigger the account recording operation.

[0118] In this embodiment, when the completion notification is transmitted, the step triggering interface of the clearing and settlement system is invoked. The identifier and content of the completion notification for the fund collection operation are input, and the interface transmits the notification to the triggering node of the account recording step. When the account recording operation is triggered, the triggering node of the account recording step receives the completion notification, reads the total amount of collected funds, the number of collected accounts, and the collection completion time from the notification, and initiates the account recording operation. The operation includes detailed fund transfer records and account balance calculations. The triggering process record includes the trigger time, notification identifier, and operation status, and is stored in the step triggering log database of the clearing and settlement system for traceability of subsequent operations.

[0119] Step S146: Based on the detailed data of fund flow and the account balance data, after completing the account recording operation, execute the fund transfer operation, transfer the collected funds according to the preset recipient information, and obtain the fund transfer processing result data. Based on the fund transfer processing result data, after the fund transfer operation is completed, execute the voucher generation operation to generate a clearing and settlement voucher containing fund flow information, account record information and transfer result information, and obtain the clearing and settlement voucher data.

[0120] In this embodiment, during fund transfer operations, preset recipient information is read. This recipient information includes account identifier, received amount, and transfer method. Based on the sum of amounts in the categorized fund data, the funds are transferred to the corresponding account according to the received amount. During the transfer process, the transfer time, transfer amount, and recipient confirmation status are recorded, generating fund transfer processing result data. During voucher generation operations, fund flow details, account balance data, and fund transfer processing result data are integrated. Following the format requirements of clearing and settlement vouchers, clearing and settlement voucher data is generated, including voucher number, generation time, fund flow summary, account balance status, and transfer result summary. After the voucher data is generated, it is digitally signed. The signature is used to verify the authenticity of the voucher. After signing, it is stored in the voucher database for subsequent auditing and querying.

[0121] Step S147: Based on the clearing and settlement voucher data and the transmission record data of each node, record the execution data of each step in the entire clearing and settlement process, and generate a clearing and settlement execution data set. The execution data includes operation start time data, operation completion time data, operation result data, and abnormal situation data.

[0122] In this embodiment, when recording execution data, the start and completion times of each step in the clearing and settlement process are read to generate operation start time data and operation completion time data; the result status of each step is read to generate operation result data; and abnormal information occurring in each step is read, including abnormal type, abnormal time, and abnormal handling method, to generate abnormal situation data. The above data is arranged in the order of steps to form a clearing and settlement execution data set. The set includes step identifiers, start times, completion times, result data, and abnormal data. The set is stored in the clearing and settlement log database for subsequent collection and analysis of execution data.

[0123] Step S150: Collect the transmission data of the payment routing dynamic adjustment path and the execution data of the clearing and settlement process adaptation strategy, and transmit the transmission data of the payment routing dynamic adjustment path and the execution data of the clearing and settlement process adaptation strategy to the evolution update node of the deep reinforcement learning co-evolution model to update the decision parameters of the deep reinforcement learning co-evolution model.

[0124] In this embodiment, when collecting transmission data, the transmission time data, data integrity data, and forwarding status data from the transmission record data set of each node are read and integrated into transmission data for dynamically adjusting the payment routing path. When collecting execution data, the operation time data, result data, and exception data from the clearing and settlement execution data set are read and integrated into execution data for the clearing and settlement process adaptation strategy. After the transmission data and execution data are integrated, they are encapsulated according to the format requirements of the data feedback receiving channel of the evolution update node. The encapsulation content includes data identifier, collection time, transmission data block, and execution data block. After encapsulation, it is sent to the evolution update node through the data transmission protocol. After receiving the data, the evolution update node parses the data content, extracts the feature values ​​required for adjusting the decision parameters, and updates the decision parameters of the deep reinforcement learning co-evolution model according to the feature values ​​and the preset parameter adjustment rules. After the parameters are updated, the model is validated. The validation content includes decision output accuracy and parameter stability. After successful validation, the updated parameters are saved for subsequent model decision-making processes.

[0125] Step S151: Extract the transmission data of the dynamically adjusted payment route from each transmission node of the target payment channel. The transmission data of the dynamically adjusted payment route includes the transmission time data, data integrity data, node response data and path congestion data of each node, and obtain the extracted transmission data.

[0126] In this embodiment, when extracting transmission data, the log storage address of each transmission node in the target payment channel is accessed, and the transmission time field, integrity identifier field, response status field, and congestion status field in the node logs are read. The contents of these fields are used as transmission time data, data integrity data, node response data, and path congestion data, respectively. During the extraction process, the data is format-converted, converting unstructured log content into structured field data. The conversion rules include field separators and data type mappings. After conversion, the data is integrated into the extracted transmission data, which includes node identifiers, transmission time data, data integrity data, node response data, and path congestion data. This data is stored in the transmission data acquisition database for subsequent data integration and transmission.

[0127] Step S152: Extract the execution data of the clearing and settlement process adaptation strategy from the clearing and settlement execution data set. The execution data of the clearing and settlement process adaptation strategy includes fund collection efficiency data, account record accuracy data, fund transfer timeliness data, and voucher generation integrity data to obtain the extracted execution data.

[0128] In this embodiment, when extracting execution data, the start and completion times of the fund collection step in the settlement execution data set are read, and the difference between the completion time and the start time is calculated as fund collection efficiency data. The balance data of the account recording step is read, and the percentage of records with a difference of zero is used as account recording accuracy data. The difference between the transfer time and the required time of the fund transfer step is read, and the percentage of records with a difference less than a threshold is used as fund transfer timeliness data. The integrity identifier of the voucher generation step is read, and the percentage of records identified as complete is used as voucher generation integrity data. The extracted execution data includes collection efficiency data, record accuracy data, transfer timeliness data, and voucher integrity data, and is stored in the execution data acquisition database for subsequent data integration and transmission.

[0129] Step S153: Classify the extracted transmission data and the extracted execution data according to their source and type, and associate and store them according to the corresponding payment route dynamic adjustment path and settlement process adaptation strategy to generate a structured data update set, and transmit it to the evolution update node of the deep reinforcement learning collaborative evolution model according to the preset data transmission format.

[0130] In this embodiment, when classifying and associating stored data, the extracted transmission data is classified according to the source of the payment channel node, and the extracted execution data is classified according to the type of clearing and settlement step. Then, the transmission data is associated with the corresponding payment route dynamic adjustment path identifier, and the execution data is associated with the corresponding clearing and settlement process adaptation strategy identifier, generating association records. Each association record includes a path identifier, a strategy identifier, a transmission data block, and an execution data block; multiple records constitute a structured data update set. During data transmission, the data update set is encapsulated according to a preset data transmission format, which includes a header identifier, data length, checksum, and data block order. After encapsulation, the data is sent to the listening port of the evolution update node via a network transmission protocol. Transmission progress and status are recorded during transmission, and confirmation information from the node is awaited upon completion of transmission.

[0131] Step S154: After receiving the data update set, the evolution update node of the deep reinforcement learning co-evolution model analyzes and processes the data update set based on the parameter adjustment logic of deep reinforcement learning, identifies the key decision factors that affect the decision-making effect, and obtains the key decision factor identification result.

[0132] For example, in step S1541: After receiving the structured data update set, the evolution update node of the deep reinforcement learning co-evolution model performs format parsing on the data in the structured data update set, extracts the data field identifiers and corresponding data values, and obtains the parsed data fields and data values.

[0133] In this embodiment, when parsing the data update set, the encapsulated format of the structured data update set is read. Following the field order and data type in the format, the field identifier and corresponding data value of each data entry are extracted. The field identifier includes a transmission data field identifier and an execution data field identifier. The data value includes transmission time data value, data integrity data value, fund collection efficiency data value, and account record accuracy data value. During the parsing process, the data value is type-validated to ensure that the data value type matches the type corresponding to the field identifier. After successful type validation, the parsed data fields and data values ​​are stored in the parsed data cache of the evolution update node for subsequent preprocessing.

[0134] Step S1542: Based on the parsed data fields and data values, and based on the parameter adjustment logic of deep reinforcement learning, the parsed data fields and data values ​​are input into the input stage of the parameter adjustment logic. Based on the parsed data fields and data values ​​received by the input stage, the input stage of the parameter adjustment logic preprocesses the data. Based on the preprocessed data, the feature analysis stage of the parameter adjustment logic performs feature analysis on the preprocessed data to identify the changing trends and correlations in the data, and obtain data trend and correlation information.

[0135] In this embodiment, when inputting data to the parameter adjustment logic, the input interface of the parameter adjustment logic is called to convert the parsed data fields and data values ​​into a format recognizable by the interface before inputting them into the input stage. During data preprocessing, the input stage performs missing value imputation and outlier handling. Missing value imputation uses the mean of data within the same field, and outlier handling uses a truncation method based on standard deviation. During feature analysis, the feature analysis stage performs trend analysis and correlation analysis on the preprocessed data. Trend analysis calculates the rate of change of data values ​​over time, and correlation analysis calculates the correlation coefficients between different data fields. The rate of change and correlation coefficients are integrated into data trend and correlation information, including field identifiers, rate of change, and correlation coefficients, and stored in the feature analysis result database of the evolution update node for subsequent impact assessment.

[0136] Step S1543: Based on data trend and correlation information, construct data impact assessment logic. Based on the data impact assessment logic and the preprocessed data, input the preprocessed data into the data impact assessment logic, calculate the impact weight related data of each data item, and obtain the impact weight data of each data item.

[0137] In this embodiment, when constructing the data impact assessment logic, calculation rules for impact weights are set based on the rate of change and correlation coefficient in the data trend and correlation information. These rules include weight coefficients for the rate of change and correlation coefficients. When calculating the impact weight-related data, preprocessed data is input into the data impact assessment logic. The logic calculates the impact weight value for each data item according to the calculation rules. The impact weight value is the sum of the rate of change multiplied by its weight coefficient and the correlation coefficient multiplied by its weight coefficient. The calculation results are integrated into the impact weight data for each data item, which includes field identifiers and impact weight values. This data is stored in the impact weight database of the evolution update node for subsequent key data item filtering.

[0138] Step S1544: Based on the influence weight data of each data item, set the filtering conditions for the data related to the influence weight, filter out the data items that meet the filtering conditions, and generate the key data items that affect the decision-making effect.

[0139] In this embodiment, when setting the filtering conditions, a filtering threshold for the influence weight values ​​is set based on the parameter adjustment experience of the deep reinforcement learning model. The threshold is the average of the influence weight values ​​of all data items multiplied by a preset multiple. When filtering key data items, the influence weight values ​​in the influence weight data of each data item are read, and data items that are greater than the filtering threshold are regarded as key data items affecting the decision-making effect. Key data items include field identifiers and influence weight values, and are stored in the key data item database of the evolution update node for subsequent classification and report generation.

[0140] Step S1545: Based on the key data items that affect the decision-making effect, classify the selected key data items, distinguish between key data items related to transmission and key data items related to execution, and obtain the classified key data items. Based on the classified key data items, generate a key data item identification report. The key data item identification report includes the name of the key data item, the corresponding influence weight related data, the data source, and the influence related description.

[0141] In this embodiment, when classifying key data items, the field identifiers of the key data items are read. Based on the prefixes in the identifiers, key data items related to transmission and those related to execution are distinguished. Key data items related to transmission have the prefix "transmission," while those related to execution have the prefix "execution." When generating a key data item identification report, the field identifiers and influence weight values ​​of the classified key data items are read. The names, data sources, and influence descriptions corresponding to the field identifiers are queried. This information is then integrated into a key data item identification report, which includes the name, influence weight value, data source, and influence description. This report is stored in the identification report database of the evolution update node for input in subsequent parameter adjustment stages.

[0142] Step S1546: Based on the key data item identification report, transmit the key data item identification report to the parameter adjustment stage of the evolution update node of the deep reinforcement learning co-evolution model.

[0143] In this embodiment, when transmitting the identification report, the input interface of the evolution update node parameter adjustment stage is called to convert the key data item identification report into a format that the interface can recognize and input it into the parameter adjustment stage. During the input process, the transmission time, report identifier, and interface status are recorded. After the transmission is completed, the system waits for the reception confirmation information from the parameter adjustment stage. The confirmation information includes the reception time and report integrity identifier, which are used for the traceability of the subsequent parameter adjustment process.

[0144] Step S155: Based on the key decision factor identification results, determine the categories and adjustment directions of the decision parameters that need to be adjusted in the deep reinforcement learning co-evolution model. Based on the categories and adjustment directions of the decision parameters that need to be adjusted and the relevant data in the data update set, locate the corresponding decision parameters in the deep reinforcement learning co-evolution model, perform parameter correction processing, and after the parameter correction is completed, synchronize the corrected decision parameters to the decision output node and interactive simulation node of the deep reinforcement learning co-evolution model, update the decision generation logic and simulation operation logic, and use the updated decision generation logic and simulation operation logic to perform verification operations on historical payment request data through the deep reinforcement learning co-evolution model to generate verification payment routing paths and clearing and settlement strategies. Compare the verification payment routing paths and clearing and settlement strategies with the corresponding historical actual execution results to complete the update of the decision parameters of the deep reinforcement learning co-evolution model.

[0145] In this embodiment, when determining the parameter category and adjustment direction, the category of the model decision parameter is matched according to the factor identifier of the key decision factor. The category includes path weight parameters, process adaptation parameters, and correlation strength parameters. The adjustment direction of the parameter is determined according to the sign of the correlation ratio; a positive ratio corresponds to the direction of parameter increase, and a negative ratio corresponds to the direction of parameter decrease. When locating and correcting parameters, the corresponding parameter storage address in the model is found according to the parameter category, the current parameter value is read, and the parameter value is corrected by combining the relevant data changes in the data update set. The correction rules include the weight coefficient of the change and the upper and lower limit thresholds of the parameter. When synchronizing parameters, the corrected parameter value is sent to the parameter update interface of the decision output node and the interactive simulation node. After receiving the data, the interface updates the decision generation logic and simulation operation logic of the node. During the verification operation, historical payment request data is read, input into the updated model, and a verification payment routing path and clearing and settlement strategy are generated. These are compared with the paths and strategies of the historical actual execution results, and the similarity ratio is calculated. If the similarity ratio is higher than the threshold, the parameter update is completed; otherwise, the above parameter adjustment process is repeated.

[0146] Furthermore, Figure 2 A schematic diagram of the hardware structure of an intelligent payment routing and clearing optimization system 100 utilizing deep reinforcement learning to implement the methods provided in the embodiments of this application is shown. Figure 2 As shown, the intelligent payment routing and clearing optimization system 100 utilizing deep reinforcement learning may include at least one processor 102 (the processor 102 may be, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, a transmission device 106 for communication functions, and a controller 108. Those skilled in the art will understand that... Figure 2The structure shown is for illustrative purposes only and does not limit the structure of the intelligent payment routing and clearing optimization system 100 utilizing deep reinforcement learning. For example, the intelligent payment routing and clearing optimization system 100 utilizing deep reinforcement learning may also include components that are more complex than... Figure 2 The more or fewer components shown, or having the same Figure 2 The different configurations shown.

[0147] The memory 104 can be used to store software programs and modules of the application software, such as the program instructions corresponding to the method embodiments described above in this application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the above-described intelligent payment routing and clearing optimization method using deep reinforcement learning. The transmission device 106 is used to acquire or send data via a network.

[0148] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

Claims

1. A method for optimizing intelligent payment routing and clearing using deep reinforcement learning, characterized in that, The method includes: Establish a real-time correspondence between key information of payment requests and clearing and settlement execution rules. The real-time correspondence is used to reflect the clearing and settlement operation standards and execution requirements corresponding to various key information of payment requests. A deep reinforcement learning co-evolution model is constructed, which includes information perception nodes, interactive simulation nodes, decision output nodes, and evolution update nodes. By analyzing the interaction between payment channel transmission data and clearing and settlement execution rules through a deep reinforcement learning collaborative evolution model, dynamic adjustment paths for payment routing and adaptation strategies for clearing and settlement processes are generated. Based on the payment routing, the path is dynamically adjusted to guide the payment request to complete the data transmission within the corresponding payment channel, and the entire clearing and settlement process is executed synchronously according to the clearing and settlement process adaptation strategy. The system collects transmission data of the payment routing dynamic adjustment path and execution data of the clearing and settlement process adaptation strategy, and transmits the transmission data of the payment routing dynamic adjustment path and execution data of the clearing and settlement process adaptation strategy to the evolution update node of the deep reinforcement learning co-evolution model to update the decision parameters of the deep reinforcement learning co-evolution model.

2. The intelligent payment routing and clearing optimization method using deep reinforcement learning according to claim 1, characterized in that, The construction of the deep reinforcement learning co-evolution model includes: Collect historical interaction data between payment channels and clearing and settlement processes. This historical interaction data includes transmission response data, resource usage data, and process connection data of payment channels under different clearing and settlement execution rules. Based on the historical interaction data, descriptive information of payment channel transmission-related data is extracted. This descriptive information includes data transmission efficiency data and transmission stability data characterizing transmission performance, as well as process compatibility evaluation data and rule adaptability evaluation data characterizing interaction features. Together, these constitute a set of descriptive data related to payment channel transmission. Based on the historical interaction data, execution detail information of clearing and settlement execution rules is extracted. This execution detail information includes data describing the fund processing status, account record status, fund transfer status, and voucher generation status at different processing stages. Together, these constitute a set of details for clearing and settlement execution rules. A deep reinforcement learning co-evolution model association mining network structure is constructed, which includes feature input nodes, interaction simulation nodes, association extraction nodes and association storage nodes. The payment channel transmission-related data description set and the clearing and settlement execution rule detail set are input into the feature input nodes of the association mining network structure for alignment and fusion processing to obtain a fused dataset for simulation analysis; Based on the fusion dataset, we can perform scenario-based interactive simulations, extract and store related data, and build model data links to complete the construction of a deep reinforcement learning co-evolution model.

3. The intelligent payment routing and clearing optimization method using deep reinforcement learning according to claim 2, characterized in that, The process of performing scenario-based interactive simulations based on fused datasets, extracting and storing related data, and building model data links to complete the construction of a deep reinforcement learning co-evolution model includes: By using the interactive simulation nodes of the aforementioned association mining network structure, and based on the fusion dataset, the interaction process between the transmission behavior of the payment channel and the processing behavior of the clearing and settlement execution rules is simulated, generating scenario-based interactive simulation results. The association extraction nodes of the association mining network structure are used to separate and extract explicit and implicit association data from the scenario-based interactive simulation results to obtain the separated and extracted association data. The association storage nodes of the association mining network structure are used to perform structured storage processing on the implicit association data in the separated and extracted association data to generate a set of implicit association data with payment channel transmission related data as the core index, clearing and settlement execution rules as the association clues, and interactive effect data as the association strength identifier. A data connection is established between the implicit association dataset and the decision output node of the deep reinforcement learning co-evolution model. A data feedback receiving channel is set at the evolution update node of the deep reinforcement learning co-evolution model. The data feedback receiving channel is used to receive the transmission data of the payment routing dynamic adjustment path and the execution data of the clearing and settlement process adaptation strategy collected subsequently.

4. The intelligent payment routing and clearing optimization method using deep reinforcement learning according to claim 1, characterized in that, The method of analyzing the interaction between payment channel transmission data and clearing and settlement execution rules through a deep reinforcement learning collaborative evolution model, and generating dynamic adjustment paths for payment routing and adaptation strategies for clearing and settlement processes, includes: The key information contained in the received payment request is parsed to generate an adaptation target description for payment routing and clearing and settlement. The key information includes data on the initiation scenario of the payment request, data on the scale of fund transfer, data on transmission timeliness requirements, and data on related business types. Based on the aforementioned adaptation target description of payment routing and clearing and settlement, the implicit correlation data set stored in the deep reinforcement learning co-evolution model is retrieved, and payment channel transmission related data items and clearing and settlement execution rule data items with high correlation to the key information contained in the received payment request are selected to obtain a matching data item set. The payment channel transmission related data and clearing and settlement execution rule data in the matching data item set are input into the information perception node of the deep reinforcement learning co-evolution model for real-time data perception and feature extraction processing to obtain the corresponding feature data. The information sensing node transmits the feature data to the interactive simulation node of the deep reinforcement learning co-evolution model. The interactive simulation node simulates the interaction process under different combinations of payment channels and different clearing and settlement processes, and evaluates and obtains the interaction effect evaluation data of each combination. Based on the combined interaction effect evaluation data, the system generates candidate paths and processes, filters for suitability, constructs dynamic adjustment logic, and generates path strategies for output.

5. The intelligent payment routing and clearing optimization method using deep reinforcement learning according to claim 4, characterized in that, The process of generating candidate paths and processes, filtering for suitability, constructing dynamic adjustment logic, and generating path strategies based on combined interaction effect evaluation data includes: Based on the combined interaction effect data, the decision output node of the deep reinforcement learning co-evolution model initially generates multiple payment routing candidate paths and corresponding clearing and settlement candidate processes, thus obtaining a set of initially generated candidate paths and candidate processes. Based on the combined interaction effect evaluation data and the correlation strength identifier in the implicit correlation data set, the adaptability analysis is performed on multiple payment route candidate paths and clearing and settlement candidate process combinations in the initially generated candidate path and candidate process set, and candidate combinations with adaptability meeting the preset threshold are selected to obtain the selected candidate combinations. For the selected candidate combinations, the dynamic adjustment logic is generated by the decision output node of the deep reinforcement learning collaborative evolution model to determine the adjustment trigger conditions and adjustment methods of payment routing under different transmission scenarios, thus obtaining the dynamic adjustment logic; Based on the adjustment trigger conditions and adjustment methods in the dynamic adjustment logic, the node connection details and transmission order of the payment route candidate paths in the filtered candidate combinations are refined to generate the payment route dynamic adjustment path; Based on the transmission characteristics of dynamically adjusted payment routing paths, the execution steps and connection nodes of the clearing and settlement candidate processes in the selected candidate combinations are synchronously matched to generate clearing and settlement process adaptation strategies that are compatible with the dynamically adjusted payment routing paths. The dynamically adjusted payment routing paths and clearing and settlement process adaptation strategies are then associated and bound to generate a complete output result containing the correspondence between paths and strategies.

6. The intelligent payment routing and clearing optimization method using deep reinforcement learning according to claim 1, characterized in that, The process of dynamically adjusting the path based on payment routing to guide payment requests to complete data transmission within the corresponding payment channel, and simultaneously executing the entire clearing and settlement process according to the clearing and settlement process adaptation strategy, includes: Extract payment channel identifier data and path node data contained in the dynamically adjusted payment routing path, determine the target payment channel and transmission node order that the payment request needs to access, and obtain target payment channel information and transmission node order information; Based on the target payment channel identifier data in the target payment channel information, a transmission connection is established between the payment request and the target payment channel. The data transmission routing guidance information is set according to the transmission node sequence information to obtain the transmission connection and routing guidance information. The key information of the payment request is integrated with the routing guidance information to generate a data packet containing transmission instructions. The data packet containing transmission instructions is then sent to the first transmission node of the target payment channel through the established transmission connection. Each transmission node of the target payment channel receives and forwards data packets containing transmission instructions in sequence according to the routing guidance information. During the forwarding process, the transmission time data and data integrity data of each node are recorded to obtain the transmission record data of each node. The execution of the clearing and settlement process is triggered by the transmission record data, and a set of full-process operation records and execution data is generated.

7. The intelligent payment routing and clearing optimization method using deep reinforcement learning according to claim 6, characterized in that, The process of triggering the clearing and settlement process based on transmission record data, and the generation of the entire process operation record and execution data set, include: When a data packet containing transmission instructions is transmitted to the terminal node of the target payment channel, the execution instruction of the clearing and settlement process adaptation strategy is triggered. Based on the execution instruction of the clearing and settlement process adaptation strategy, and in accordance with the execution steps included in the clearing and settlement process adaptation strategy, the fund collection operation is first performed to collect the fund data corresponding to the payment request and classify and organize it to obtain the classified and organized fund data. Based on the classified and organized fund data, the account recording operation is performed to record the fund flow details and account balance data in accordance with the requirements of the clearing and settlement execution rules to obtain the fund flow details data and account balance data. Based on the detailed data of fund flow and the data of account balance, after the account recording operation is completed, the fund transfer operation is performed. The collected funds are transferred in a targeted manner according to the preset recipient information to obtain the fund transfer processing result data. Based on the fund transfer processing result data, after the fund transfer operation is completed, the voucher generation operation is performed to generate a clearing and settlement voucher containing fund flow information, account record information and transfer result information to obtain clearing and settlement voucher data. Based on the clearing and settlement voucher data and the transmission record data of each node, the execution data of each step in the entire clearing and settlement process is recorded to generate a clearing and settlement execution data set. The execution data includes operation start time data, operation completion time data, operation result data, and abnormal situation data.

8. The intelligent payment routing and clearing optimization method using deep reinforcement learning according to claim 1, characterized in that, The data collected includes the transmission data of the payment routing dynamic adjustment path and the execution data of the clearing and settlement process adaptation strategy. This data is then transmitted to the evolution update node of the deep reinforcement learning co-evolution model to update the decision parameters of the model, including: The transmission data of the dynamically adjusted payment route is extracted from each transmission node of the target payment channel. The transmission data of the dynamically adjusted payment route includes the transmission time data, data integrity data, node response data and path congestion data of each node, and the extracted transmission data is obtained. The execution data of the clearing and settlement process adaptation strategy is extracted from the clearing and settlement execution data set. The execution data of the clearing and settlement process adaptation strategy includes fund collection efficiency data, account record accuracy data, fund transfer timeliness data and voucher generation completeness data to obtain the extracted execution data. The extracted transmission data and the extracted execution data are classified according to their source and type, and associated with each other based on the corresponding payment route, dynamic adjustment path and settlement process adaptation strategy, to generate a structured data update set, and then transmitted to the evolution update node of the deep reinforcement learning collaborative evolution model according to the preset data transmission format. Based on the updated data set, key decision factors are identified, parameters are corrected and synchronized, and models are validated and updated.

9. The intelligent payment routing and clearing optimization method using deep reinforcement learning according to claim 8, characterized in that, The process of identifying key decision factors, synchronizing parameter corrections, and updating the model for validation based on the updated data set includes: After receiving the data update set, the evolution update node of the deep reinforcement learning co-evolution model analyzes and processes the data update set based on the parameter adjustment logic of deep reinforcement learning, identifies the key decision factors that affect the decision-making effect, and obtains the key decision factor identification result. Based on the key decision factor identification results, the categories and directions of decision parameters that need to be adjusted in the deep reinforcement learning co-evolution model are determined. Based on the categories and directions of decision parameters that need to be adjusted and the relevant data in the data update set, the corresponding decision parameters in the deep reinforcement learning co-evolution model are located, and parameter correction processing is performed. After the parameter correction is completed, the corrected decision parameters are synchronized to the decision output node and interactive simulation node of the deep reinforcement learning co-evolution model, and the decision generation logic and simulation operation logic are updated. Using the updated decision generation logic and simulation operation logic, the deep reinforcement learning co-evolution model is used to perform verification operations on historical payment request data to generate verification payment routing paths and clearing and settlement strategies. The verification payment routing paths and clearing and settlement strategies are compared with the corresponding historical actual execution results to complete the update of the decision parameters of the deep reinforcement learning co-evolution model.

10. An intelligent payment routing and clearing optimization system utilizing deep reinforcement learning, characterized in that, It includes a processor and a readable storage medium storing a program that, when executed by the processor, implements the intelligent payment routing and clearing optimization method using deep reinforcement learning as described in any one of claims 1-9.

Images