Multi-region payment gateway intelligent routing dispatch system and method

Abstract

The present application relates to the technical field of distributed system resource scheduling, and particularly relates to a multi-region payment gateway intelligent routing scheduling system and method, and the present application proposes the following scheme: when there is a payment request to be processed, the regional scheduling behavior information corresponding to each region is obtained; based on the regional scheduling behavior information and the regional feedback information, a regional state folding graph is constructed; for the target region corresponding to the payment request to be processed, a plurality of candidate payment channels are determined based on the regional state folding graph, and the current observed state of the candidate payment channels in the target region is folded, disturbed and stripped to obtain the intrinsic available state corresponding to each candidate payment channel; according to the intrinsic available state, the target payment channel corresponding to the payment request to be processed is determined, and the payment request to be processed is scheduled to the target payment channel. The folding interference of the cross-regional scheduling behavior on the channel state can be identified, the real available state of the channel is restored, the channel misjudgment is avoided, and the overall scheduling efficiency and success rate of the multi-region are improved.

Description

Technical Field

[0001] This invention relates to the field of distributed system resource scheduling technology, and in particular to a multi-regional payment gateway intelligent routing scheduling system and method. Background Technology

[0002] Currently, in the scheduling architecture of multi-region payment gateways, payment requests from different geographical or logical regions typically share the underlying payment channel resources. Most existing scheduling schemes are designed based on a static resource contention model, which assumes that payment channels have an objective and uniform availability state. Each region makes routing decisions by reading this global state and combining it with its local load conditions. Under this model, the main focus is on channel capacity occupancy, using load balancing algorithms to prevent any single channel from becoming overloaded.

[0003] However, in real-world business scenarios, the calls to the same underlying payment channel from different regions do not merely manifest as a static resource competition relationship. High-intensity scheduling behavior in Region A dynamically alters how the payment channel's status is displayed in other regions through mechanisms such as shared risk control links, clearing priority queues, authorization queue reordering, and failure tag reinjection. For example, a large number of orders in Region A may trigger sensitive thresholds in the underlying risk control model, leading to stricter risk control reviews for subsequent orders in Region B. Alternatively, failed transactions in Region A, once marked, may affect the channel's success rate statistics from Region B's perspective. This phenomenon causes channels that were originally available in Region B to be misjudged as unavailable or excessively downgraded due to interference from Region A—a phenomenon known as "state folding interference." Existing scheduling technologies cannot identify this cross-regional, implicit state folding interference, causing locally optimal scheduling decisions to disrupt the global state perception order, resulting in overall scheduling imbalance across multiple regions and reducing the processing efficiency and success rate of the payment system.

[0004] To address the above issues, this application presents a multi-regional payment gateway intelligent routing and scheduling system and method. Summary of the Invention

[0005] To address the technical problem in existing technologies that fail to identify the folding interference of cross-regional scheduling behavior on channel status, leading to overall scheduling imbalance in multiple regions, this application proposes a multi-regional payment gateway intelligent routing scheduling method. This method constructs a regional status folding graph and removes folding disturbances from the observed status, thereby achieving accurate identification of the intrinsic available status of the channel and stable maintenance of the global scheduling order.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A multi-region payment gateway intelligent routing and scheduling method, wherein each region corresponds to multiple payment channels, and each payment channel corresponds to different regional feedback information in different regions. The regional feedback information is used to characterize the local availability, risk, and responsiveness of the payment channel in the corresponding region. The method includes:

[0008] When there are pending payment requests, obtain the corresponding regional scheduling behavior information for each region;

[0009] Based on regional scheduling behavior information and regional side feedback information, a regional state folding diagram is constructed;

[0010] For the target area corresponding to the payment request to be processed, based on the area state folding map, multiple candidate payment channels are determined, and the current observation state of the candidate payment channels in the target area is folded and perturbed to obtain the intrinsic available state corresponding to each candidate payment channel.

[0011] Based on the intrinsic availability status, the target payment channel corresponding to the pending payment request is determined, and the pending payment request is scheduled to the target payment channel.

[0012] The method for determining the regional side feedback information includes:

[0013] Obtain historical payment processing data for historical payment requests in different regions and payment channels. The historical payment processing data includes at least the region identifier, payment channel identifier, processing result identifier, failure reason identifier, result return delay information, and retry processing information.

[0014] The historical payment processing data is classified and aggregated according to the region identifier and payment channel identifier to obtain the set of historical feedback events for each payment channel in each region.

[0015] Based on the distribution of successful processing events, failed processing events, risk interception events, timeout events, and retry events in each historical feedback event set, the local availability, local risk, and local responsiveness of the corresponding payment channel in the corresponding region are determined.

[0016] Based on the local availability, local risk, and local responsiveness, regional feedback information for the corresponding payment channel in the corresponding region is generated.

[0017] The regional scheduling behavior information is used to characterize the call intensity, transaction type distribution, failure clustering, and scheduling bias of each payment channel within the corresponding region. This regional scheduling behavior information is obtained through the following methods:

[0018] Obtain the scheduling results of each region for each payment channel within the preset statistical time window;

[0019] The scheduling result information is aggregated by region based on the region identifier, and the previous scheduling result corresponding to each payment channel is determined in each region according to the payment channel identifier.

[0020] Based on the previous scheduling results for each payment channel, generate regional scheduling behavior information for the corresponding area.

[0021] If a region does not have a previous scheduling result for a payment channel within a preset statistical time window, the region's scheduling behavior information is generated in the following ways:

[0022] Based on the regional state folding diagram of the previous scheduling cycle, a reference region corresponding to the region is determined, and the previous scheduling result of the corresponding payment channel in the reference region is extracted.

[0023] The previous scheduling result is corrected by regional folding to generate a substitute scheduling result for the corresponding payment channel in the reference area, and corresponding regional scheduling behavior information is generated based on the substitute scheduling result.

[0024] The method for constructing the region state folding map includes:

[0025] Based on the regional scheduling behavior information corresponding to each region, the scheduling effect characteristics of each region on each payment channel within the current scheduling cycle are determined. The scheduling effect characteristics include at least the call intensity effect characteristics, transaction type effect characteristics, failure clustering effect characteristics, and scheduling bias effect characteristics.

[0026] Based on the regional feedback information of each payment channel in each region, the status display characteristics of each payment channel in each region are determined. The status display characteristics include at least local availability display characteristics, local risk display characteristics, and local responsiveness display characteristics.

[0027] Based on the correlation between the scheduling characteristics of each region and the status display characteristics of each payment channel in each region, the cross-regional folding interference relationship between the scheduling behavior of different regions and the status display of each payment channel in other regions is determined.

[0028] Based on the cross-regional folding interference relationship between regions, a regional state folding diagram is constructed. The regional state folding diagram is used to characterize the folding effect of the scheduling behavior of different regions on the state manifestation of each payment channel in different regions. The nodes of the regional state folding diagram are regional nodes, and the edges are the inter-regional folding interference edges corresponding to the payment channels.

[0029] The method for determining the target region includes:

[0030] Obtain the region identification information corresponding to the payment request to be processed. The region identification information includes at least one of the following: payment initiation region identifier, payment terminal home region identifier, account home region identifier, currency identifier, and merchant access region identifier.

[0031] Based on the region identification information, the region affiliation result corresponding to the payment request to be processed is determined;

[0032] The region corresponding to the region attribution result is determined as the target region.

[0033] The method for determining the candidate payment channel includes:

[0034] Based on the target area, obtain multiple payment channels within the area corresponding to the target area, as an initial payment channel set;

[0035] Obtain the current observation status of each payment channel in the initial payment channel set in the target area, and based on the area state folding map, determine the folding interference of other areas on each payment channel in the target area;

[0036] Based on the current observation status of each payment channel and the folding interference situation, determine the channel suppression status of each payment channel;

[0037] Payment channels whose channel suppression status meets the preset suppression conditions are eliminated, and the remaining payment channels are identified as the multiple candidate payment channels.

[0038] The folding disturbance stripping includes:

[0039] Extract the local availability observation component, local risk observation component, and local responsiveness observation component from the current observation status of each candidate payment channel in the target area;

[0040] Extract the inter-regional folding interference edges from other regions pointing to the target region and corresponding to the corresponding candidate payment channel from the regional state folding diagram, and determine the external folding disturbance component of the corresponding candidate payment channel in the target region based on the folding interference intensity corresponding to each inter-regional folding interference edge.

[0041] By deducting the corresponding external folding disturbance components from the local availability observation component, the local risk observation component, and the local responsiveness observation component, respectively, the local availability component, local risk component, and local responsiveness component of the corresponding candidate payment channel in the target area are obtained;

[0042] Based on the local availability component, the local risk component, and the local responsiveness component, the intrinsic availability status of the corresponding candidate payment channel in the target area is determined.

[0043] The method for determining the target payment channel includes:

[0044] The intrinsic availability status score of each candidate payment channel is generated based on the intrinsic availability status.

[0045] Based on the intrinsic availability score of each candidate payment channel, the candidate payment channel with the highest intrinsic availability score is determined as the target payment channel, and the pending payment request is scheduled to the target payment channel.

[0046] A multi-regional payment gateway intelligent routing and scheduling system, the system comprising:

[0047] The regional information generation module is used to generate regional scheduling behavior information corresponding to each region and regional side feedback information of each payment channel in each region. The regional scheduling behavior information is used to characterize the call intensity, transaction type distribution, failure clustering and scheduling bias of each payment channel in the corresponding region. The regional side feedback information is used to characterize the local availability, risk and responsiveness of the payment channel in the corresponding region.

[0048] The folded graph construction module constructs a regional state folded graph based on the regional scheduling behavior information and the regional side feedback information. The nodes of the regional state folded graph are regional nodes, and the edges are inter-regional folding interference edges corresponding to the payment channels. The regional state folded graph is used to characterize the folding impact of the scheduling behavior of different regions on the state manifestation of each payment channel in different regions.

[0049] The candidate channel analysis module is used to determine the target area corresponding to the pending payment request when there is a pending payment request, determine multiple candidate payment channels corresponding to the target area based on the area state folding map, and perform folding perturbation stripping on the current observation state of each candidate payment channel in the target area to obtain the intrinsic available state corresponding to each candidate payment channel.

[0050] The routing and scheduling module is used to determine the target payment channel corresponding to the pending payment request based on the intrinsic availability status of each candidate payment channel, and to schedule the pending payment request to the target payment channel.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] This invention constructs a regional state folding diagram, enabling the quantitative identification of the folding interference relationship between scheduling behaviors in different regions and the cross-regional state manifestation of payment channels. The method is no longer limited to the traditional static resource competition perspective, but delves into the level of state interpretation rights. By stripping external folding disturbance components from the current observed state, it obtains the intrinsic availability state, reflecting the true availability of the channel. Scheduling decisions based on this intrinsic availability state effectively avoid misjudgments, incorrect avoidance, or excessive deweighting of channel states caused by high-intensity scheduling in other regions, maintaining the stability of the overall multi-regional scheduling order and significantly improving the global processing success rate and resource utilization efficiency of the payment gateway. Attached Figure Description

[0053] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0054] Figure 1 An exemplary application scenario diagram provided for an embodiment of the present invention;

[0055] Figure 2 This is a flowchart illustrating the intelligent routing and scheduling method for multi-regional payment gateways provided in an embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0057] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0058] The intelligent routing and scheduling method for multi-regional payment gateways described in this application is applicable to payment processing scenarios with multiple regional business entry points, significant mutual influence between regional payment behaviors, and regionally differentiated payment channel status. More specifically, this application is particularly applicable to cross-border e-commerce platforms, global digital service platforms, multi-country merchant platforms, regional wallet aggregation platforms, and comprehensive payment networks that simultaneously provide payment services to different clearing jurisdictions. In such application environments, although payment requests appear on the surface as transactions initiated against a single region, from the perspective of the underlying payment chain's operational mechanism, calls to the same payment channel from different regions are not isolated but often implicitly coupled through shared authorization links, risk control links, clearing resources, retry mechanisms, and channel-side feedback mechanisms. Therefore, the status feedback of the same payment channel in different regions is inconsistent, and its success rate, risk sensitivity, and response characteristics in a certain region are often indirectly affected by the scheduling behavior of other regions.

[0059] It should be noted that this application does not presuppose a fixed binding relationship between payment channels and regions, nor does it require different regions to share the exact same channel resource pool. The focus of this application is that when multiple regions concurrently call several payment channels during the same period, the regional state of the payment channel is no longer solely determined by the transaction results within that region. Instead, it may exhibit a result with folded perturbations due to cross-regional scheduling. In other words, in multi-regional payment scenarios, those skilled in the art typically hope to use the channel performance observed in a particular region to determine whether the channel is suitable to continue handling subsequent requests from that region. However, in actual engineering operations, the premise that what is seen within a region is the true state within that region often does not hold true. Especially in scenarios involving high-frequency concurrent payments, cross-regional synchronous marketing activities, short-term migration of regional traffic, dynamic tightening of channel-side risk control models, and long asynchronous confirmation chains, the channel state currently observed in a particular region may have already been mixed with perturbations from previous scheduling behaviors in other regions.

[0060] In existing technologies, channel routing for multi-region payment requests typically involves candidate filtering and priority ranking based on channel rates, historical success rates, response latency, or regional admission rules. In practical deployments, those skilled in the art often further incorporate methods such as failure retries, regional weight adjustments, downgrading of abnormal channels, and local window statistics to improve payment success rates in a single region. However, these approaches largely assume that each region's perception of channel status is relatively independent; that is, they assume that the success rate, failure rate, and latency distribution obtained within a region are sufficient to represent the true scheduling value of that channel in that region. This approach can usually meet basic usage requirements in situations with small regional scales, weak inter-regional coupling, or relatively stable channel loads. However, when multiple regions simultaneously contend for channel resources, and the channel status changes in tandem with regional call behavior, relying solely on local observations within a region for routing decisions can easily lead to a situation where the locally optimal scheduling behavior of one region gradually alters the state perception of the same payment channel in other regions. This results in the latter observing a channel status that no longer reflects the true state corresponding to their own transaction behavior, but rather a mixed result superimposed with external regional disturbances.

[0061] It is important to emphasize that the core logic of this application lies in the fact that the key contradiction in multi-regional payment networks is not merely whether a certain region is monopolizing more channel processing resources, nor simply whether a certain channel is currently in a high success rate state. Rather, it lies in the fact that scheduling behaviors in different regions cause cross-regional propagation and folding at the state feedback level of payment channels, leading to deviations between local observations and the actual local state. In other words, the reason why many routing decisions in existing technologies still exhibit instability in multi-regional scenarios is not because there are insufficient statistical dimensions or the parameters are not updated quickly enough, but because the state basis used for scheduling decisions has itself been contaminated by cross-regional behavior. If these external folding disturbances cannot be identified and removed first, simply continuing to optimize rates, rank success rates, or compensate for failures in the disturbed state often only yields superficially more complex, but essentially still unstable, scheduling results.

[0062] Based on the above understanding, this application does not follow the existing approach of simply evaluating routing based on results within a region. Instead, it starts from the formation mechanism of channel states in multi-regional payment scenarios, incorporating regional scheduling behavior and the state manifestation of payment channels in each region into a unified analysis framework. This further identifies the folding interference relationships formed between regions around channel states and constructs a regional state folding graph based on this. Different regions are no longer regarded as independent parallel routing units, but as coupled participants that may mutually alter the channel state manifestation results. The state of the same payment channel in different regions is no longer simply understood as a direct mapping of a unified global state in each region, but as a regional manifestation result affected by the combined effects of local transaction contributions and external regional disturbances. Furthermore, this application performs folding disturbance stripping on the current observed state of candidate payment channels in the target region, separating out the intrinsic usable state that better characterizes the true scheduling value of the channel in the target region. This ensures that subsequent payment channel routing is no longer based on aliased states, but on the de-scratched local true states.

[0063] It should also be noted that the application is not limited to traditional cross-border credit card payment environments, but can also be extended to local wallets, transfer gateways, aggregated acquiring links, regional clearing channels, and payment systems with asynchronous confirmation or multi-level risk control review processes. The technical approach proposed in this application is generally applicable as long as any of the following characteristics exist in such payment systems: First, different regions share some payment channel resources or share channel-side state feedback links; second, the same payment channel exhibits significantly different local availability, risk, or responsiveness in different regions; third, the scheduling results in one region have a continuous impact on the subsequent channel status judgments in other regions; fourth, the local statistical results may deviate from the local actual state due to the superposition of cross-regional behaviors. For such scenarios, this application does not require the division of payment regions to be fixed in advance to a specific form such as country, site, clearing jurisdiction, or merchant cluster. As long as mutually distinguishable regional scheduling units can be formed, and the calls to payment channels by each region cause perceptible state coupling, the technical principles of this application are valid.

[0064] refer to Figure 1 , Figure 1 This is an exemplary application scenario diagram provided for an embodiment of this application.

[0065] like Figure 1As shown, multiple regional scheduling units process payment requests around multiple payment channels. Different regions can initiate scheduling requests for corresponding payment channels based on their respective business entry points, payment object distribution, merchant access relationships, and channel access conditions. In the illustrated example, regions A, B, C, and D can be understood as mutually distinct regional scheduling units, while payment channel one, payment channel two, and payment channel three can be understood as payment processing channels that different regions can choose to use.

[0066] It should be noted that, Figure 1 This is only used to illustrate the application relationship of multiple regions and multiple payment channels. It does not mean that any region necessarily has a fixed one-to-one correspondence with all payment channels, nor does it mean that each region has completely consistent access capabilities or scheduling permissions for each payment channel.

[0067] It should be further clarified that in practical applications, a certain region may not correspond to all payment channels. For example, some payment channels may only be open to specific regions, some may only support transaction requests under specific currencies, specific merchant types, or specific settlement links, and some payment channels, although logically associated with a certain region, may not participate in the current scheduling of that region during a specific period due to risk control restrictions, capacity constraints, channel maintenance, or business strategy adjustments. In this application, this situation can be understood as: each region corresponds to its own set of payment channels within the region, which represents the range of payment channels in the corresponding region that are eligible for scheduling, have access conditions, or are candidate selections under the current business conditions. In other words, in this application, each region corresponds to multiple payment channels, and it is not required that all regions share the exact same set of payment channels. Rather, it emphasizes that each region can form local scheduling behavior within its corresponding channel range, and this local scheduling behavior may still affect other regions through the status feedback link of the payment channels.

[0068] In this context, although the payment channel sets corresponding to different regions may only partially overlap, be locally monopolized, or dynamically change, the technical solution proposed in this application remains applicable as long as there are state transfer, feedback coupling, resource competition, or risk linkage relationships between different regions surrounding at least some payment channels. This application does not rely on a strictly fully connected topology between multiple regions and multiple payment channels, but rather focuses on the engineering reality that the scheduling behavior of a region over its payment channels may change the state manifestation of those payment channels in other regions, thereby causing the local availability, risk, or responsiveness subsequently observed in other regions to be subject to folding perturbations introduced by the scheduling behavior of external regions.

[0069] like Figure 2As shown, this embodiment provides a multi-region payment gateway intelligent routing scheduling method. The method is applied to a multi-region payment gateway system. Each region corresponds to multiple payment channels. Each payment channel corresponds to different regional side feedback information in different regions. The regional side feedback information is used to characterize the local availability, risk and responsiveness of the payment channel in the corresponding region.

[0070] Specifically, this embodiment breaks away from the traditional understanding that payment channels possess an objective and uniform global availability state. In an architecture where multiple regions share underlying channel resources, the performance of the same payment channel often varies drastically across different regions. For example, a payment channel might exhibit high availability in region A due to high currency matching, but in region B, it might exhibit high latency due to a long clearing path. Therefore, the regional feedback information described in this embodiment refers to a set of data observed from the perspective of a specific region, reflecting the local characteristics of the channel within that region. This information includes at least three dimensions: local availability, characterizing the probability of the channel successfully processing payment requests within that region; risk, characterizing the probability of triggering risk control interception and fraud warnings; and responsiveness, characterizing time efficiency characteristics such as processing latency and timeout rate.

[0071] A method for intelligent routing and scheduling of multi-regional payment gateways specifically includes the following steps:

[0072] Step S100: When there are pending payment requests, obtain the regional scheduling behavior information corresponding to each region.

[0073] Specifically, regional scheduling behavior information is the root cause of cross-regional state folding. In multi-region scheduling scenarios, high-intensity scheduling behavior in one region, such as initiating a large number of requests in a short period of time, or specific types of transaction behavior, such as high-frequency failure retries, will have a subtle impact on other regions through shared risk control links, clearing queues, and other underlying mechanisms. This step aims to capture these dynamic behavioral characteristics of each region in real-time or near real-time, including but not limited to call intensity, transaction type distribution, and failure clustering.

[0074] Step S200: Construct a regional state folding diagram based on regional scheduling behavior information and regional side feedback information.

[0075] Specifically, traditional solutions only focus on resource capacity utilization, neglecting the interference of state perception. The regional state folding graph constructed in this embodiment is a topological structure diagram used to characterize the folding impact of scheduling behavior in different regions on the state manifestation of each payment channel in different regions. In the graph, nodes represent regions, and edges represent folding interference relationships between regions. Folding interference refers to the fact that the scheduling behavior of region A changes the interpretation of the same channel's state by region B. For example, high-concurrency requests in region A trigger underlying risk control, causing the payment channel to appear as high-risk from the perspective of region B; this distortion of state perception is folding interference. By constructing this graph, the system can explicitly quantify and record this implicit cross-regional influence relationship.

[0076] Step S300: For the target area corresponding to the payment request to be processed, based on the area state folding map, determine multiple candidate payment channels, and perform folding perturbation stripping on the current observation state of the candidate payment channels in the target area to obtain the intrinsic available state corresponding to each candidate payment channel.

[0077] Specifically, this step aims to address the problem of how to restore the true state. In the presence of folding interference, the channel state observed in the target area, i.e., the current observation state, is often contaminated, containing interference components from other areas.

[0078] Understandably, the current observed state refers to the apparent data directly measured by the system, such as the current high failure rate; external folding disturbances are interference components injected from other regions, calculated based on the regional state folding diagram, such as the increase in failure rate caused by the refeedback from region A; the intrinsic usability state is the true usability that the channel should have in the target region after removing interference. By performing folding disturbance removal operations, the system can eliminate false unavailability factors, restore the intrinsic capability of the payment channel, and thus avoid misjudging or over-weighting of usable channels due to interference from other regions.

[0079] Step S400: Based on the intrinsic availability status, determine the target payment channel corresponding to the pending payment request, and schedule the pending payment request to the target payment channel.

[0080] Specifically, after obtaining the intrinsic availability state that reflects the true capabilities, the system no longer makes decisions based on contaminated observation states, but instead scores and ranks them according to the intrinsic availability state. The system selects the channel with the optimal intrinsic availability state score as the target payment channel for scheduling. This decision-making mechanism ensures the accuracy of the scheduling results, effectively maintains the stability of the overall scheduling order across multiple regions, and avoids global imbalances caused by locally optimal decisions.

[0081] Through the above solution, this embodiment identifies cross-regional interference by constructing a regional state folding diagram and restores the intrinsic usable state of the channel by stripping away the disturbance, thus solving the channel misjudgment and scheduling imbalance problems caused by state folding interference in the prior art and significantly improving the global processing success rate of the payment gateway.

[0082] In one example, this application embodiment further elaborates on the process of generating regional feedback information and the method for locating the target region.

[0083] In the first aspect, regarding the method for determining the regional side feedback information, this embodiment adopts the following approach:

[0084] Obtain historical payment processing data for historical payment requests in different regions and payment channels. The historical payment processing data includes at least the region identifier, payment channel identifier, processing result identifier, failure reason identifier, result return delay information, and retry processing information.

[0085] Specifically, historical payment processing data consists of raw log records accumulated during system operation. To support subsequent folding interference analysis, this data must possess fine-grained characteristics. Region identifiers distinguish the geographical or logical region of the data source; payment channel identifiers distinguish the underlying payment links, such as channel A-Visa link, channel B-UnionPay link; processing result identifiers record whether the transaction was successful or failed; failure reason identifiers further refine the specific type of failure, such as insufficient balance, risk control interception, network timeout, etc.; result return latency information records the time difference from initiation to receiving a response; retry processing information records whether the transaction underwent retry scheduling. This data typically exists in the form of a log stream, and the system maintains a sliding time window to capture this data, such as the most recent hour or the most recent 24 hours, to ensure data timeliness.

[0086] The historical payment processing data is classified and aggregated according to the region identifier and payment channel identifier to obtain the historical feedback event set of each payment channel in each region.

[0087] Specifically, the classification and aggregation process is the process of transforming massive amounts of raw logs into structured data.

[0088] In this embodiment, the region identifier and payment channel identifier are used as a joint primary key to map the dispersed historical payment processing data to the corresponding set. Based on the distribution of successful processing events, failed processing events, risk interception events, timeout events, and retry events in each historical feedback event set, the local availability, local risk, and local responsiveness of the corresponding payment channel in the corresponding region are determined.

[0089] Specifically, this step involves quantifying and extracting characteristic indicators from the original event distribution. Local availability can be characterized by calculating the percentage of successfully processed events in the total number of events. For example, if the percentage of successfully processed events in a set is 98%, it indicates that the channel has high local availability in that region. Local risk is characterized by statistically analyzing the frequency of risk-blocking events. If the percentage of risk-blocking events is abnormally high, it indicates that the channel faces high risk control pressure in that region. Local responsiveness is comprehensively characterized by statistical values ​​of result return latency information and the percentage of timeout events and retry events. The lower the latency and the fewer the timeout retries, the better the local responsiveness.

[0090] It is important to emphasize that these indicators can be understood as local indicators, meaning they only reflect the performance of the channel within a specific region, rather than the global uniform state.

[0091] Based on the local availability, local risk, and local responsiveness, regional feedback information for the corresponding payment channel in the corresponding region is generated.

[0092] Specifically, when extracting local availability from a set of historical feedback events, it's not simply a matter of counting successful events. Instead, events included in the statistics need to be standardized to ensure that the subsequent availability results accurately reflect the payment channel's capacity within the target area. In this embodiment, historical feedback events can be stratified according to their closure level. For example, events with a clear final state are considered valid completion samples, while events still in pending, unconfirmed, or intermediate return states are considered undetermined samples. Non-standard termination events such as abnormal interruptions, repeated callbacks, and link breaks are separately marked. Subsequently, in the local availability calculation, valid completion samples are used as the primary statistical basis, and the proportion of successfully processed events in these samples is used as the basic availability component. Simultaneously, the basic availability component is adjusted for stability by considering the concentration of failed processing events within adjacent time slices. The reason is that even if a payment channel has a high overall success rate within a statistical window, a significant clustering of failure events within a short period usually indicates that the channel is experiencing temporary instability, link fluctuations, or tightening of channel-side restrictions in that region. Without correcting this phenomenon, a channel that is temporarily unstable but has a high historical average can easily be misjudged as a consistently high-availability channel. Therefore, the local availability referred to in this embodiment is essentially a comprehensive reflection of the channel's actual capacity and stability in handling payment requests within the corresponding region, rather than a mechanical mapping of a single success rate value.

[0093] It should be noted that the concentration of failed events within adjacent time slices can be determined through time slice division and intra-slice failure density extraction. Specifically, the statistical window used to calculate local availability can be divided into multiple consecutive time slices in chronological order. The length of each time slice can be set according to the payment request density and the frequency of channel status changes, for example, 1 minute, 3 minutes, or 5 minutes. For each time slice, the number of events belonging to valid completion samples, the number of failed events, and the failure reason category corresponding to each failed event are counted. The ratio of the number of failed events to the number of valid completion samples in that time slice is taken as the intra-slice failure density. Specifically, the stability correction method can be implemented using existing sliding window anomaly detection, time series fluctuation penalty, or event clustering correction algorithms, as long as it can adjust the basic availability component downward based on the intra-slice failure density and its changes in adjacent time slices. This application does not specify or elaborate on this.

[0094] Furthermore, in determining localized risk, this embodiment does not simply interpret risk as "the number of failures," but rather focuses on identifying whether the payment channel exhibits a rejection trend with risk control characteristics within the corresponding region. This is because ordinary failure events may stem from network jitter, user cancellations, insufficient account balances, or proactive restrictions imposed on specific transaction patterns by the payment channel or related risk control links. Mixing all failures of different natures in the statistics could easily obscure the true source of risk pressure. Therefore, this embodiment first refines and categorizes the reasons for failures in the historical feedback event set, identifying events related to risk review, anti-fraud interception, rule rejection, identity verification failure, and abnormal transaction patterns as risk interception events. Then, it statistically analyzes their distribution ratio in the total events, their frequency of consecutive occurrences, and their degree of clustering under specific transaction types. If the proportion of risk interception events for a certain payment channel increases within a certain region and maintains a high density across several consecutive time slices, it can be inferred that the channel has entered a highly sensitive state in that region, meaning that the transaction structure, account characteristics, or payment behavior patterns within that region are more likely to trigger the risk assessment mechanism on the channel side. Therefore, local risk is not an abstract label, but a state quantity that reflects the payment channel’s risk sensitivity, rejection tendency and risk control pressure level to transaction requests in a specific area. Its role is to provide a risk-side explanation basis independent of success rate for subsequent state manifestation analysis.

[0095] Regarding the determination of local responsiveness, this embodiment should not rely solely on the average return latency for a single-value judgment. Instead, it should comprehensively consider the latency distribution pattern, timeout event density, and retry event propagation to more accurately characterize the response efficiency and link smoothness of the payment channel in the corresponding region. For example, the result return latency information can be segmented and statistically analyzed to extract the median latency, long-tail latency ratio, and timeout event percentage, distinguishing between two different scenarios: "overall slow response" and "a small number of extremely slow requests dragging up the average." Then, combined with the frequency of retry events in adjacent calls, it can be determined whether the payment channel has issues with unstable initial requests and the need for repeated submissions to complete transactions. This is because in actual payment scenarios, although some channels may have a relatively high final success rate, if they generally rely on multiple retries or frequently experience long-tail returns, it will significantly increase the uncertainty of the scheduling link within the region and amplify the impact on subsequent scheduling behavior. In this embodiment, the lower the latency, the shorter the long tail, the fewer the timeouts and the less frequent the retries, the higher the local responsiveness of the corresponding channel in the region; conversely, it indicates that there is a significant response lag or link vulnerability in the region.

[0096] The median latency refers to the latency value in the middle of the result return latency of each valid completed sample within the corresponding statistical window, arranged in ascending order. For example, if there are 501 valid completed samples, and the return latency of the 251st sample is 460ms, then the median latency is 460ms. The long-tail latency ratio refers to the proportion of valid completed samples whose result return latency exceeds the long-tail latency limit among the total number of valid completed samples. The long-tail latency limit can be determined based on the latency quantile of valid completed samples in the same region within the payment channel over the past 7 days, or it can be determined in conjunction with the response warning value agreed upon in the channel access agreement. For example, if the 90th quantile latency over the past 7 days is 1800ms and the channel response warning value is 2000ms, then 1800ms can be taken as the long-tail latency limit. If there are 1000 valid completed samples within the statistical window, and 86 of them exceed 1800ms, then the long-tail latency ratio is 8.6%. The timeout event percentage refers to the proportion of events that fail to receive a valid response within the channel timeout period within the statistical window to the total number of events entering the responsiveness statistics. The channel timeout period can be configured by the payment gateway or determined by the channel access protocol. For example, if the channel protocol stipulates that a timeout occurs if a final state is not returned within 8 seconds, and there are 600 events entering the responsiveness statistics within the statistical window, of which 12 events have not returned a valid response after 8 seconds, then the timeout event percentage is 2%.

[0097] Specifically, the generated regional feedback information can be a multi-dimensional vector or a structured data object containing quantized values ​​of the three dimensions mentioned above.

[0098] Secondly, regarding the method for determining the target area, this embodiment adopts the following approach:

[0099] Obtain the region identification information corresponding to the payment request to be processed. The region identification information includes at least one of the following: payment initiation region identifier, payment terminal home region identifier, account home region identifier, currency identifier, and merchant access region identifier.

[0100] Specifically, when a new payment request enters the system, it doesn't inherently carry a clear "target region" label. Instead, it carries various region-related attributes. The payment initiation region identifier is usually derived from resolving the request's source IP address; the payment terminal's region identifier is determined by the POS machine or mobile device's registration location; the account's region identifier is determined by the user's bank card opening bank or account registration location; the currency identifier reflects the currency used in the transaction, such as USD or EUR, with different currencies often corresponding to different clearing regions; and the merchant access region identifier indicates the merchant's registration or operating location. This information may point to different regions.

[0101] Based on the region identification information, determine the region attribution result corresponding to the payment request to be processed; and determine the region corresponding to the region attribution result as the target region.

[0102] Specifically, the system needs to determine a final target region from multi-dimensional region identification information based on a preset region attribution strategy. This strategy can be priority-based, for example, prioritizing the account's region identifier; if the account's region identifier is missing, the payment initiation region identifier is used. Alternatively, it can be weight-based, for example, comprehensively considering the payment initiation region, account's region, and merchant access region, calculating the weight score for each region, and determining the region with the highest score as the target region. Furthermore, currency identifiers can be used for verification; for example, if the determined region does not support the currency, the region is adjusted. The significance of determining the target region lies in the fact that all subsequent state folding analysis and intrinsic state restoration are based on the perspective of that target region. Only with accurate localization can the observed state from the perspective of that region be accurately identified, and folding interference from other regions be eliminated, thereby making correct scheduling decisions.

[0103] It should be noted that the specific weight score can be calculated using a multi-factor weighted matching algorithm, and this application does not impose a unique limitation on its specific mathematical form.

[0104] In yet another example, this application embodiment further elaborates on the specific methods for obtaining regional scheduling behavior information and the cold start processing mechanism for data missing scenarios.

[0105] In the first aspect, the regional scheduling behavior information is used to characterize the intensity of the corresponding region's calls to each payment channel within the region, the distribution of transaction types, the clustering of failures, and the scheduling bias.

[0106] Specifically, call intensity reflects the influence of a region on a particular payment channel within a specific time period, which can be quantified by the number of concurrent requests or the total number of requests per unit time. Transaction type distribution characterizes the composition of requests, such as whether it is dominated by high-frequency, small-amount virtual goods transactions or low-frequency, large-amount physical goods transactions. Different transaction type distributions have drastically different triggering mechanisms for the underlying risk control model. Failure clustering is used to identify the spatiotemporal clustering effect of failures; for example, consecutive failures within a short period often indicate a sudden change in channel status. Scheduling bias reflects the region's tendency when choosing multiple channels, such as whether it always prioritizes the channel with the lowest fee. These characteristics collectively constitute a profile of the sources of disturbance to the channel status in this region.

[0107] The regional scheduling behavior information is obtained through the following methods:

[0108] Obtain the scheduling results of each region for each payment channel within the preset statistical time window.

[0109] The scheduling result information is aggregated by region based on the region identifier, and the previous scheduling result corresponding to each payment channel is determined in each region according to the payment channel identifier.

[0110] Based on the previous scheduling results for each payment channel, generate regional scheduling behavior information for the corresponding area.

[0111] Specifically, the collected data should be understood as the original behavioral basis for reconstructing how a region used each payment channel in recent scheduling activities, the resulting distribution of outcomes, and whether this distribution reflects local biases and instability trends. Regional scheduling behavior is highly time-sensitive; the business structure, request density, access object type, and channel selection preferences of the same region often differ at different times. Mixing scheduling results from earlier periods with current-period behavior in statistical analysis can easily mask the emerging regional behavioral characteristics with long-term averages, leading to a regional state folding diagram that is more biased towards historical static relationships and fails to reflect the real-time disturbance relationships between regions around payment channels during the current scheduling period. Therefore, in this application, the preset statistical time window is not arbitrarily chosen but is set based on the payment business update rate, channel state change frequency, and regional traffic fluctuation rhythm. For example, for scenarios with rapid daily traffic fluctuations, intensive marketing activities, or frequent changes in regional access volume, the preset statistical time window can be set to the most recent 15 minutes to the most recent 2 hours; for scenarios with relatively smooth payment behavior changes and longer channel state switching cycles, it can be set to the most recent 6 hours to the most recent 24 hours. The time window must ensure that each payment channel in the region has sufficient scheduling records for analysis, while avoiding the inclusion of outdated records in the statistics, which would dull the characteristics of current scheduling behavior.

[0112] In the actual acquisition process, the scheduling result information may include at least the region identifier, payment channel identifier, request initiation time, scheduling occurrence time, transaction type identifier, payment request source identifier, scheduling result identifier, and necessary abnormal notes. The scheduling result identifier includes not only success and failure, but can also be further subdivided into categories such as success after suspension, failure after suspension, active rejection, risk control interception, timeout pending, retry successful, and retry failed.

[0113] It is understandable that regional scheduling behavior is not solely constituted by a single routing action at a particular moment, but rather by a series of related actions, including which channel the request is assigned to, whether the channel accepts it, what processing result is returned under what latency conditions after acceptance, and whether repeated scheduling or subsequent compensation paths are triggered. If only a single record of a request being sent to a channel is obtained from the routing log, it is often impossible to determine whether the call constitutes valid scheduling, let alone whether the region's use of the channel is stable acceptance or passive probing. Therefore, in specific extraction, records from multiple sources can be concatenated based on payment request identifiers or transaction association identifiers, and then a scheduling chain for a single request can be formed according to chronological order. Subsequently, each scheduling chain can be divided into scheduling result units that can be used for behavioral analysis. A scheduling result unit can be understood as a scheduling action with closed-loop results initiated by a region for a payment channel in a certain time slice. It includes not only the fact that the channel was called, but also whether the call resulted in a final valid result, whether it triggered multiple retries, whether it was accompanied by risk rejection, and whether timeout tailing occurred. In cases involving parallel calls or fallback channel switching, scheduling priority markers can be retained to identify whether a region has a priority dependency on a particular payment channel within that time window.

[0114] Furthermore, regional aggregation is not simply about putting records with the same regional identifier into the same set. Instead, it aims to recover, at the regional level, the order of recent use of different payment channels, the status of the most recent results, and the business context of the most recent call within that region.

[0115] Understandably, regional scheduling behavior information essentially reflects a region's current usage attitude towards each payment channel. The most direct representation of this usage attitude is not the cumulative number of times a channel has been called historically, but rather the scheduling conclusion reached when the region last interacted with the channel in the most recent timeframe. If the last scheduling resulted in normal acceptance, low-latency return, and no triggering of a compensation path, it indicates that the region is still maintaining a sustainable usage status for the channel. If the last scheduling resulted in failure, accumulated suspension, or frequent retries without a valid result, it indicates that the region's usage relationship with the channel has changed.

[0116] After collecting the original scheduling results, the data needs to be aggregated first using the region identifier as the primary key, and then further subdivided within each region according to the payment channel identifier, thus forming multiple region-channel local result sequences. Based on this, each local result sequence needs to be reordered according to the scheduling occurrence time, removing duplicate write-backs, abnormal reverse order, invalid placeholders, and dirty data without result closure, ensuring that the record at the end of the sequence truly represents the most recent effective scheduling status of that region for that payment channel within that time window.

[0117] In some optional implementations, the determination of the previous scheduling result for each payment channel can employ time sequence verification, result loop verification, and anomaly coverage verification. Time sequence verification prioritizes determining the most recent record based on the scheduling execution time within the same region and payment channel's result sequence, rather than solely relying on log write time. This is because real-world engineering scenarios involve delayed receipt writes, asynchronous backups, and delayed retry log arrivals; directly using write time as the sorting criterion could easily misjudge an earlier but later-written receipt as the previous scheduling result. Result loop verification addresses situations such as successful suspension followed by a successful retry, or multiple callbacks merging into a single final state. It requires tracing the final state of the same request chain; only records reflecting the final scheduling conclusion can be considered the previous scheduling result for that channel. Anomaly coverage verification addresses complex scenarios such as repeated scheduling within a short period, fallback switching, and concurrent multi-channel testing, performing coverage checks on conflicting last records. For example, if a region initiates two scheduling requests for payment channel 1, the first request is suspended, and the second request triggers compensation and successfully switches to another channel within a short period of time, simply retaining the first suspension record will not accurately reflect the region's actual usage of payment channel 1 at the end of the current window. In this case, it is necessary to combine information such as subsequent compensation actions, whether the original channel has completed the result loop, and whether it has been explicitly abandoned by the region's actions to correct the final result, so that it is closer to the region's actual latest scheduling relationship.

[0118] Furthermore, the key to generating regional scheduling behavior information is not to output the previous scheduling results as the behavior information, but to construct an overall scheduling profile of the region in the current statistical period based on the previous scheduling results of each payment channel.

[0119] Understandably, regional scheduling behavior information is not a single result of a single channel, but rather a structured representation of how multiple payment channels within the region have been allocated, biased, imbalanced, and whether local abnormal clusters have formed in the recent past. The previous scheduling result was chosen as the core basis for generating behavioral information because it most closely approximates the region's current actual scheduling state and reflects the region's latest usage conclusions of different payment channels in the recent final stages. However, without further parallel analysis and correlation extraction of the final results of multiple channels, it is still impossible to obtain the regional-level call intensity, transaction type distribution, failure clustering, and scheduling bias.

[0120] In this embodiment, the previous scheduling results of all payment channels within a region can be collected first, and then cross-analyzed according to result type, proximity of occurrence time, consistency of transaction type, and channel switching relationship to extract regional behavioral characteristics. For example, if the previous scheduling results of most payment channels in a certain region are concentrated in a short period of time within a preset statistical time window, and a large number of records belong to the same transaction type, it indicates that there is obvious concentrated calling behavior in that region during that time period; if the last results of multiple channels are all failures or timeouts, and these results occur in adjacent time slices, it indicates that there is a recent clustering of failures in that region; if the previous scheduling result of a certain channel repeatedly appears in the region's last sequence and maintains a high usage frequency, it reflects that the region has formed a scheduling bias towards that channel. The regional scheduling behavior information generated in this way is actually a linkage summary of the last interaction results of multiple regions and channels, rather than a simple mapping of a single result.

[0121] It's important to note that call intensity can be comprehensively characterized by the distribution density of the last results of each payment channel within the current statistical window and the compactness of adjacent call intervals, rather than solely by the total number of calls. Even if a region has a high cumulative number of calls over a longer time window, it doesn't necessarily mean it's highly active in the current scheduling cycle; conversely, if multiple previous scheduling results appear concentrated in a short time slice, it better reflects the region currently experiencing high call pressure. Transaction type distribution can be categorized by combining the transaction type identifiers corresponding to the previous scheduling results of each channel, extracting which types of business requests the region tends to send to which channels recently. For example, are local wallet requests more concentrated in certain channels, and are cross-currency transactions mainly handled by a few channels? This provides a basis for subsequently judging the region's channel usage structure. Failure clustering is not just about the number of failure records, but rather the way failure results cluster in time and channels; that is, if failures only appear sporadically in several unrelated time periods, they are more likely random events, while if failures continuously fall on the last results of multiple channels within a short time window, it indicates a stronger abnormal propagation trend at the regional level recently. Scheduling bias can be determined by comparing the relative concentration of the last scheduling results of each payment channel and the channel's sustained acceptance performance in short-cycle adjacent reference segments. If a region consistently prioritizes a certain type of request to a particular channel in multiple rounds of local behavior, it can be concluded that the region has a relatively stable biased usage relationship with that channel. After the above analysis, the regional scheduling behavior information is no longer a few isolated labels, but rather reflects the dynamic scheduling structure formed by the region around multiple payment channels in the current cycle.

[0122] It is important to emphasize that in practical applications, scenarios involving cold starts or missing data are frequently encountered. For example, when a new payment channel has just been launched, or when a certain region has no scheduling records within the statistical time window, the system cannot obtain valid previous scheduling results. To address this issue, this embodiment provides a cold start handling mechanism.

[0123] If a region does not have a previous scheduling result for a payment channel within a preset statistical time window, the region's scheduling behavior information is generated in the following ways:

[0124] Based on the regional state folding diagram of the previous scheduling cycle, a reference region corresponding to the region is determined, and the previous scheduling result of the corresponding payment channel in the reference region is extracted.

[0125] Specifically, the reference region is not randomly selected, but rather based on graph structure similarity matching. The system analyzes the region state folding graph of the previous scheduling cycle to find the region most similar to the current region in terms of topological structure. For example, if the current region C exhibits low node connectivity and its main associated channel is channel X in the graph, the system will search for another region D in the graph with similarly low connectivity and its main associated channel is also channel X, and identify it as the reference region. This graph structure-based matching logic ensures that the reference region and the current region have inherent similarities in behavioral patterns, thereby improving the credibility of the substitute data. Subsequently, the system extracts the previous scheduling result of channel X in reference region D as the initial reference value.

[0126] The previous scheduling result is corrected by regional folding to generate a substitute scheduling result for the corresponding payment channel in the reference area, and corresponding regional scheduling behavior information is generated based on the substitute scheduling result.

[0127] Specifically, directly reusing data from the reference area may introduce bias because there are inherent differences between different areas. The core logic of the correction lies in subtracting the background differences between areas.

[0128] In some optional implementations, the differences between the reference region D and the current region C in basic indicators such as historical average success rate and average response latency can be calculated.

[0129] For example, if the historical average success rate of region D is 95%, while that of region C is 90%, the difference is -5%. Assuming the success rate component in the extracted previous scheduling result is 0.9, the success rate component in the corrected substitute scheduling result is adjusted to 0.85. This correction generates a virtual substitute scheduling result that conforms to the characteristics of the current region. Based on this substitute result, regional scheduling behavior information can be generated, thus enabling the construction of a reasonable regional state folding diagram even in the event of data loss, ensuring the continuity and robustness of the scheduling system.

[0130] In yet another example, embodiments of this application elaborate on a method for constructing a region state folding diagram.

[0131] Methods for constructing region state folding diagrams include:

[0132] Step S501: Based on the regional scheduling behavior information corresponding to each region, determine the scheduling effect characteristics of each region on each payment channel within the current scheduling cycle. The scheduling effect characteristics include at least the call intensity effect characteristics, transaction type effect characteristics, failure aggregation effect characteristics, and scheduling bias effect characteristics.

[0133] Specifically, regional scheduling behavior information alone is insufficient to directly participate in subsequent folding relationship determination. It is also necessary to further integrate scattered records, discrete labels, and local results into "scheduling action characteristics" that can reflect the recent scheduling attitude of the region.

[0134] Understandably, the purpose of the regional state folding diagram is not to depict the static fact of whether a region has called a certain payment channel, but rather the intensity, transaction structure, failure patterns, and channel bias that a region continuously exerts on the payment channel within the current scheduling cycle. Without elevating the original behavioral information to action characteristics, even if state changes in other regions are observed later, it is difficult to determine whether these changes originate from occasional calls from the source region or from persistent behavioral pressure already formed by the source region within the scheduling cycle. Therefore, in this step, it is necessary to further organize the regional scheduling behavior information into action representations that are comparable and transferable, so that the recent action direction, strength, and continuity of the same region on different channels can be clearly characterized.

[0135] It should be noted that the call intensity feature is not a simple count of the number of calls, but rather reflects the call density, call concentration, and short-term push trend of a specific payment channel within the current scheduling cycle. The transaction type feature is not simply a summary of transaction tags, but rather reflects whether the composition of transactions, risk types, and business request patterns recently input into the channel in the region tend to be concentrated. The failure clustering feature is not simply the number of failures in the general sense, but rather characterizes whether failures are continuous in time, concentrated in channels, and form a local chain reaction in specific transaction types. The scheduling bias feature characterizes whether the region has recently been continuously directing a certain type of request or a large proportion of requests to a few channels, thereby changing the distribution of the channel's load distribution.

[0136] In this embodiment, the previous scheduling result, adjacent reference segments of the previous scheduling result, and associated transaction types, failure reasons, retry trajectories, and channel switching traces for each payment channel within the current scheduling period can be extracted on a regional basis. Then, a local effect record of the region on each payment channel is generated. Regarding the formation of the call intensity effect characteristics, not only is the number of times the channel is triggered by the region within the current scheduling period retained, but further analysis is also conducted to determine whether the temporal distribution of these scheduling actions is compact, whether they are concentrated in a short time segment, and whether the interval between adjacent scheduling actions is continuously shortening.

[0137] For example, the current scheduling cycle can be divided into several consecutive time slices, such as 5 minutes per slice, covering the most recent 30 minutes. If more than 60% of the scheduling results for a certain region on a certain payment channel are concentrated in the most recent two time slices, and the time interval between the most recent three schedulings gradually shortens, then the call intensity characteristic of that channel in that region can be identified as high-pressure calling. If the scheduling results are relatively evenly distributed throughout the entire cycle, then it can be identified as stable calling. The 60% is not a fixed limit and can be set according to the historical fluctuation level of the region. For example, first calculate the average call concentration of each channel in the region in the previous few scheduling cycles, and then take the average plus 10% to 20% as the discrimination threshold. If the regional calls of a certain cross-border merchant platform have a high concentration under normal circumstances, then the discrimination threshold can be set to about 70%, otherwise it can be set to about 55%. To assess the impact of transaction types, the transaction types accepted or attempted to be accepted by the channel within the current scheduling period can be categorized according to preset mapping rules, such as standard consumer transactions, high-value transactions, cross-currency transactions, duplicate submissions, account-sensitive transactions, and high-risk transactions. Then, it can be observed whether one or more types of transactions consistently dominate the input to the channel in that region. If a channel recently accepted multiple transactions belonging to the high-risk cluster category within that region, and these transactions persist in adjacent time slices, then the transaction type impact characteristic can be marked as high-risk clustered input. For the impact of failure clusters, a joint analysis of the continuity and correlation of failure results is required. For example, if in the most recent 10 local scheduling results, more than 6 are failures, with at least 4 occurring in adjacent time slices, or if failures are mainly concentrated in the same transaction type and on the same channel switching link, then it can be considered that the region has formed a continuous failure impact characteristic on the channel. If the current period length is set to 30 minutes, the condition "at least 4 failures occurring within adjacent time slices" can be specifically defined as more than 4 failures occurring within the last 10 minutes. If the period is extended to 2 hours, the condition can be adjusted accordingly to more than 5 failures occurring within the last 20 to 30 minutes. Regarding scheduling bias characteristics, we can observe whether the allocation of requests to multiple payment channels in the current period is balanced. If a certain payment channel continuously undertakes the majority of similar requests in the region, and this channel is consistently prioritized in the last adjacent reference segment, it can be marked as a biased receiving channel, and the scheduling bias characteristic of this channel in the region can be determined as persistent bias.

[0138] Step S502: Based on the regional feedback information of each payment channel in each region, determine the status display characteristics of each payment channel in each region. The status display characteristics include at least local availability display characteristics, local risk display characteristics, and local responsiveness display characteristics.

[0139] Specifically, the characteristics of local availability are not simply the success rate, but rather whether the channel has been stable in the target area recently, whether there has been a sharp drop, and whether there has been a phased narrowing of the capacity. The characteristics of local risk are not simply the percentage of risk control interception, but rather whether the risk review has become stricter recently, whether risk rejection is concentrated on a certain type of transaction, and whether the increase in risk is continuous. The characteristics of local responsiveness include not only the speed of return latency, but also whether the latency has started to lengthen, whether timeouts show a continuous accumulation, and whether retries have become the normal compensation path.

[0140] In this embodiment, a state observation segment can be established for each payment channel-region combination. The state observation segment takes the regional side feedback information as the core and combines it with the historical feedback event set of the payment channel in the region within the current scheduling cycle to extract the recent changes in local availability, local risk and local responsiveness.

[0141] To assess the local availability characteristics, we can first observe the proportion of successfully processed events in the effective closed-loop events within the current scheduling period, and then compare this proportion with the corresponding proportion in the adjacent previous reference period. If the current period shows a significant decrease and the downward trend is maintained within two consecutive time slices, the local availability characteristics of the payment channel in this region can be marked as acceptance contraction; if the success rate remains stable and the failure events do not expand continuously, it can be marked as stable acceptance.

[0142] For example, if a payment channel's success rate in region B was 97% in the previous reference period, drops to 91% in the current period, and remains below 92% in the last two 5-minute time slices, then the channel's local availability in region B can be considered to have shifted from stable to contracted. A decrease of 5% to 6% can serve as an example trigger range. This range can be determined by first statistically analyzing the natural fluctuation range of the channel's success rate over several historical periods in that region. If the fluctuation is usually within 2%, then a continuous decrease exceeding 5% is considered a significant change in status.

[0143] To identify localized risk characteristics, a comprehensive assessment can be made based on the proportion of risk interception events, the concentration of risk failure causes, and the sequential occurrence of risk events over time. If a channel in a certain region previously maintained a low proportion of risk interception events, but now experiences a continuous accumulation of risk events, and these events primarily correspond to a certain type of transaction or account characteristic, then the localized risk characteristics can be marked as an increase in risk.

[0144] For example, if the risk interception rate rises from 2 percent in the previous period to 7 percent in the current period, and more than 5 of the last 10 related events are similar high-sensitivity transactions, then the local risk in the region can be considered to have entered a high-sensitivity review state.

[0145] To determine the characteristics of local responsiveness, we can look at the median level of the return delay, whether the long-tail delay has extended, whether timeout events occur consecutively, and whether the number of retry links has increased. If the median return delay increases from the original 2 to 3 seconds to more than 5 seconds, and there are more than 2 timeouts in recent events accompanied by successful retries, it indicates that the local responsiveness of the channel in this region is no longer stable and has entered a tailed response or compensation response state.

[0146] Step S503: Based on the correlation between the scheduling function characteristics of each region and the status display characteristics of each payment channel in each region, determine the cross-regional folding interference relationship caused by the scheduling behavior of different regions on the status display of each payment channel in other regions.

[0147] Specifically, the regional state folding diagram represents an attributable chain of cross-regional state influences, rather than a general regional correlation. Cross-regional folding interference essentially refers to a directional, time-dependent, and difficult-to-explain correspondence between the recent scheduling patterns of a source region on a certain payment channel and the state manifestation changes of the same payment channel in the target region. Only when these conditions are met can it be considered that the scheduling behavior of the source region does not merely occur simultaneously with the state changes of the target region, but truly participates in the formation of the target region's channel state manifestation results. Therefore, in this step, it is necessary to establish correlation paths at three levels: region-to-region, channel-to-channel, and effect feature-to-state manifestation feature, to determine whether a certain region has caused folding interference to the availability, risk, or responsiveness of the same payment channel in other regions within the current scheduling cycle. In other words, this step is not a general data correlation comparison, but rather an interference identification process targeting "who influenced whom, through what effect features influenced what state manifestation, and whether this influence has a sustainable technical explanation."

[0148] In this embodiment, a candidate interference analysis chain can be constructed for each source region-target region-payment channel combination. Taking a certain payment channel as a fixed object, the call intensity effect characteristics, transaction type effect characteristics, failure aggregation effect characteristics, and scheduling bias effect characteristics of the source region on that channel are first read. Then, the local availability manifestation characteristics, local risk manifestation characteristics, and local responsiveness manifestation characteristics of the target region on the same channel are read. Subsequently, based on the preset effect-manifestation mapping rules, it is determined whether there is a valid interference path between the two. In this application, the mapping rules can be established based on the underlying mechanism of the payment link, rather than being arbitrarily set.

[0149] For example, when the source region's transaction type characteristics on a certain channel exhibit high-risk clustered input, and its call intensity characteristics are simultaneously high-pressure calls, while the target region's local risk manifestation characteristics on that channel show risk elevation in a subsequent short observation segment, and the high-risk input to the target region's corresponding channel does not increase synchronously, then "high-risk high-pressure input from the source region - risk elevation in the target region" can be identified as a valid candidate folding interference path. This relationship can be explained in engineering terms by shared risk control models, shared rule thresholds, or shared risk profile caching. That is, a large number of high-risk requests are imported into the source region in a short period, causing the underlying risk control link to tighten its review strategy for that payment channel, thus resulting in a higher interception rate for subsequent requests from the target region even if the transaction structure does not change significantly. For example, if the failure aggregation characteristic of the source region on a certain channel is characterized by continuous failures, and these failures are accompanied by retry accumulation or fallback switching, then if the local responsiveness of the target region on that channel subsequently shows increased latency, increased timeouts, or frequent retries, and the target region's own call intensity remains stable, then the "continuous failure of the source region - response tailing of the target region" can be identified as another type of candidate folding interference path. Technically, this path can be explained by shared authorization queues, shared compensation links, or channel-side degradation protection mechanisms. That is, continuous anomalies in the source region trigger the channel to enter protection or degradation mode, causing the return links of other regions on the same channel to be forced to delay. To avoid accidental accompaniments being misjudged as folding interference, time-related constraints and self-exclusion constraints can be added to each candidate interference path. The time-related constraint can be set to occur within a preset observation window after the target region's status changes after the source region's influence characteristics are enhanced, such as within five, ten, or fifteen minutes; this observation window can be set based on the typical delays of risk control propagation and queue propagation on the payment channel side. The self-exclusion constraint requires that the corresponding influence characteristics of the target region itself on the same channel do not undergo synchronous enhancement in the same direction, or even if they do, they are insufficient to explain the current state change on their own. If both constraints are satisfied, the candidate path can be confirmed as a valid cross-regional folding interference relationship.

[0150] For example, the determination of cross-regional folding interference relationships can be achieved using the following logic: The system detects that the intensity of calls from region A to channel X is characterized as "high pressure," and the transaction type is characterized as "high-risk clustering." Simultaneously, the system detects an abnormal increase in the local risk manifestation characteristics of region B to channel X, i.e., a sudden increase in the risk control interception rate. If region B's own scheduling behavior does not change significantly at this time, the system can determine that region A's high-intensity, high-risk scheduling behavior "folds" and interferes with region B's perception of channel X's state. This interference is achieved through the underlying shared risk control model: a large number of high-risk transactions from region A trigger the sensitive threshold of the underlying risk control model, causing the model to adopt a stricter review strategy for subsequent transactions from region B, thereby changing the risk manifestation of channel X from region B's perspective.

[0151] For example, the failure aggregation effect of region A on channel X manifests as "continuous failures," which may cause the underlying authorization queue of channel X to enter protection mode or degradation mode, thereby leading to "high latency" or "high timeout rate" in the local responsiveness observed in region B. This interference is achieved through a shared authorization queue mechanism. By analyzing this spatiotemporal association in a large amount of historical data, the system can quantify and determine the folding interference relationship between different regions.

[0152] Step S504: Based on the cross-regional folding interference relationship between regions, construct a regional state folding diagram. The regional state folding diagram is used to characterize the folding effect of the scheduling behavior of different regions on the state display of each payment channel in different regions. The nodes of the regional state folding diagram are regional nodes, and the edges are inter-regional folding interference edges corresponding to the payment channels.

[0153] Specifically, while a single interference record can explain "which area might affect which area" on a specific channel at a given moment, in actual scheduling, the target area faces multiple payment channels, multiple external areas, and various types of state manifestation changes. Without a unified graph structure to centrally represent these relationships, subsequent state descrambling on the target area would require searching external interference records one by one. This is not only costly but also makes it difficult to identify, as a whole, which external areas are the main sources of interference, which payment channels are the main folding carriers, and which type of interference is dominant in the current scheduling cycle. Therefore, it is necessary to treat the region as an indexable core node, encompassing the cross-regional folding interference relationships formed around specific payment channels as directed edges, with each edge carrying the corresponding channel identifier, interference type, and interference intensity information.

[0154] Understandably, when it is necessary to determine whether the current observation state of a candidate payment channel in the target area is contaminated by external regions, it is only necessary to retrieve the folded interference edge between the regions pointing to the target area and corresponding to the payment channel in the graph to quickly determine the potential source of external disturbance and its degree of influence.

[0155] In this embodiment, when constructing the regional state folding graph, a regional node can be established for each region. The regional node should at least contain the region's identifier, regional behavior update time, local channel coverage, and a summary of scheduling action characteristics related to the current scheduling period. This summary information is retained in the node so that when searching for interference sources on the graph later, it is possible to not only know which region the node corresponds to, but also to know the approximate scheduling posture of the region in the current period. Subsequently, for each cross-regional folding interference relationship confirmed in step S503, an inter-regional folding interference edge is generated, pointing from the source region node to the target region node. This edge is not an abstract edge, but a directed edge bound to a specific payment channel. That is, multiple different interference edges can be formed between the same pair of regions due to different payment channels; even the same payment channel may simultaneously carry one or more of availability folding, risk folding, and responsiveness folding. For ease of subsequent use, the edge can contain the following information: associated payment channel identifier, interference manifestation type, interference intensity level, source region action characteristic summary, target region state manifestation summary, interference effective time range, and a timeout flag used to determine whether the edge is still valid. For example, if region A forms a risky folding interference edge to region C on payment channel two, this edge can contain information such as "payment channel two, risk escalation, strong interference, high-risk clustering input on the source side, increased risk interception on the target side, valid in the current period". If, in a subsequent update period, this interference relationship no longer satisfies the time-related constraint or the target region itself has sufficient local explanatory evidence, the validity marker of this edge can be changed to a decay state or an invalid state.

[0156] It should be understood that the graph in this application is dynamically updated. As the scheduling behavior of each region changes, the weights of the edges fluctuate in real time, and even the existence of the edges may change. When region A stops high-voltage scheduling, the weights of the edges that interfere with region B will gradually decay until they disappear. This dynamic graph structure can capture the complex state entanglement relationships between multiple regions in real time, providing a solid topological model support for accurately removing disturbances and restoring the intrinsic state of the channel in subsequent steps.

[0157] In yet another example, embodiments of this application elaborate in detail the screening logic for candidate payment channels, the specific algorithm for folding perturbation stripping, and the final determination method for target payment channels.

[0158] In the first aspect, regarding the method for determining candidate payment channels, this embodiment adopts the following approach:

[0159] Based on the target area, obtain multiple payment channels within the corresponding area as an initial payment channel set.

[0160] Specifically, the initial payment channel set includes all logically available payment channel resources within the target area, regardless of their current observation status, to ensure the comprehensiveness of channel resources.

[0161] Obtain the current observation status of each payment channel in the initial payment channel set in the target area, and based on the area state folding map, determine the folding interference of other areas on each payment channel in the target area.

[0162] Specifically, the current observation status is the apparent data directly measured by the system, such as a 25% failure rate for a certain channel in the target area. The folding interference situation is obtained by querying the constructed regional state folding graph, such as whether there are high-weight interference edges pointing to the target area in the graph, and the weight value of the edges.

[0163] Based on the current observation status of each payment channel and the folding interference situation, the channel suppression status of each payment channel is determined.

[0164] Specifically, channel suppression status is a comprehensive evaluation index used to characterize whether a channel is currently in a "not available" or "extremely restricted" state. The system calculates this status through preset suppression judgment logic. For example, if a channel has an extremely high observation failure rate, such as exceeding 90%, and the folding interference situation shows that the external interference intensity is low, it indicates that the channel itself is faulty, and its suppression status is marked as "strong suppression"; if the observation failure rate is high, but the external interference intensity is also extremely high, it indicates that the channel may have been "falsely affected," and its suppression status may be marked as "weak suppression" or "pending."

[0165] Payment channels whose channel suppression status meets the preset suppression conditions are eliminated, and the remaining payment channels are identified as the multiple candidate payment channels.

[0166] Specifically, preset suppression conditions can be set according to business needs. For example, a suppression threshold can be set, and when the suppression status index exceeds the threshold, the channel is determined to be unusable.

[0167] Specifically, the suppression threshold can be determined based on the distribution of suppression status index within the same region and the same payment channel over a historical statistical period. The suppression status index can be normalized to a value between 0 and 1, with a higher value indicating that the payment channel is less suitable to continue accepting payment requests from the current region. For example, historical scheduling samples with clear processing results within the past 30 days can be selected. Samples that were ultimately accepted normally without triggering retries, timeouts, or risk interceptions are considered acceptable samples, while samples that ultimately failed, experienced consecutive retries that failed, timed out, or triggered channel-side rejection are considered suppressed samples. Then, the 95th quartile of the suppression status index corresponding to the acceptable samples and the 10th quartile of the suppression status index corresponding to the suppressed samples are calculated, and the midpoint between the two is used as the suppression threshold. For example, if the 95th quartile of the acceptable samples is 0.58 and the 10th quartile of the suppressed samples is 0.72, then the suppression threshold can be set to 0.65.

[0168] Secondly, regarding the folding disturbance stripping, this embodiment achieves it in the following way:

[0169] Extract the local availability observation component, local risk observation component, and local responsiveness observation component from the current observation status of each candidate payment channel in the target area.

[0170] Specifically, these three components correspond to the channel's observations in the three dimensions of availability, risk, and responsiveness.

[0171] In this embodiment, a status observation segment can be extracted first, focusing on the most recent effective observation window of the candidate payment channel in the target area. This observation window can be set to five to thirty minutes before the current scheduling time, or, based on the channel's receipt timeliness characteristics, to the most recent twenty, fifty, or one hundred effective closed-loop events. If a payment channel is a high-frequency call channel, it is more appropriate to prioritize the method of extracting by the number of events to ensure that the status observation segment can cover a sufficient number of recent behaviors. If a payment channel is not frequently called in the area, it is more appropriate to prioritize the method of extracting by time to ensure that effective observations are not impossible due to insufficient events. Subsequently, the percentage of successfully processed events, the continuity of effectively completed events, the distribution of abnormal interruptions, and the stability of the acceptance are extracted from the status observation segment, and combined with the corresponding results in the previous reference segment to generate a local availability observation component. If the successful closed loops in the current segment remain continuous, abnormal interruptions do not accumulate, and the acceptance results do not show a continuous decline, the local availability observation component can be identified as a stable acceptance state. If the successful closed loops decrease in several consecutive events, and the failure events change from a scattered distribution to an adjacent distribution, it can be identified as a shrinking acceptance state. Simultaneously, it is necessary to extract the recent distribution of risk interception events, rule rejection events, account-sensitive events, and identity verification anomaly events from the state observation segments. This distribution, combined with the concentration trend of these events in transaction type, their continuous trend over time, and their upward trend in adjacent windows, generates local risk observation components. If risk-type events occur only occasionally and do not form a continuous accumulation within the current window, the local risk observation component can remain as normal risk. If risk-type events consistently occur in the last ten to twenty events, and their frequency increases from low to high compared to the previous window, it can be identified as a risk-increased state. For local responsiveness observation components, it is necessary to extract result return latency, long-tail return events, timeout events, and retry-closed-loop events from the same state observation segment, observing whether there is an overall increase in latency, an increase in tail latency, or a reliance on retries for closed-loop events in the most recent calls. If the return time remains within the normal range of the region's historical data, and timeout events occur sporadically without forming a chain, the local responsiveness observation component can be identified as a stable response. If the median return time is consistently higher than the normal level of the channel in the region over the past few cycles, or if more than two out of the last ten calls have timed out and triggered compensated retries, it can be identified as a trailing response or a compensated response state.

[0172] Furthermore, the local availability observation component, local risk observation component, and local responsiveness observation component do not exist only in the form of labels such as "stable acceptance state," "risk escalation state," or "tailing response state." Instead, they are first formed by the event distribution characteristics of the candidate payment channel in the target area within the current observation window, and then the state is interpreted according to the interval or level where the observation is located.

[0173] In this embodiment, a status observation segment can be extracted first, focusing on the most recent effective observation window of the candidate payment channel in the target area. This observation window can be set to five to thirty minutes before the current scheduling time, or, based on the channel's acknowledgment timeliness characteristics, to the most recent twenty, fifty, or one hundred effective closed-loop events. If a payment channel is a high-frequency call channel, the method of extracting by the number of events is preferred to ensure that the status observation segment covers a sufficient number of recent behaviors. If a payment channel is not frequently called in the area, the method of extracting by time is preferred to avoid distortion of the observation segment due to insufficient event count. Subsequently, the percentage of successfully processed events, the continuity of effectively completed events, the distribution of abnormal interruptions, and the stability of acceptance are extracted from the status observation segment to form local availability observations. The recent distribution, continuity, and type concentration trends of risk interception events, rule rejection events, account sensitive events, and identity verification abnormal events are extracted to form local risk observations. The result return delay, long-tail return events, timeout events, and closed-loop events after retry are extracted to form local responsiveness observations. For each observation, a preset interval mapping rule can be used to convert it into a corresponding observation component. The local availability observation component characterizes the observation interval in which the channel's carrying capacity falls within the current observation window; the local risk observation component characterizes the observation interval in which the risk review pressure falls within the current observation window; and the local responsiveness observation component characterizes the observation interval in which the link return efficiency falls within the current observation window. After completing the interval mapping, based on the interval in which each observation component falls, the local availability observation component is further labeled as stable carrying capacity, fluctuating carrying capacity, or contracted carrying capacity; the local risk observation component is further labeled as normal risk, risk escalation, or high-sensitivity review; and the local responsiveness observation component is further labeled as stable response, tailing response, or compensated response.

[0174] Extract the inter-regional folding interference edges from other regions pointing to the target region and corresponding to the corresponding candidate payment channel from the regional state folding diagram, and determine the external folding disturbance component of the corresponding candidate payment channel in the target region based on the folding interference intensity corresponding to each inter-regional folding interference edge.

[0175] Specifically, the system searches for all edges pointing to the target region in the regional state folding graph. Assume there exists an edge in the graph pointing from region A to the target region, and the payment channel associated with this edge is candidate channel X. The system reads the weight of this edge; for example, a weight of 0.1 indicates that the scheduling behavior of region A has caused the risk control interception rate of channel X in the target region to increase by 10%. This 10% is the risk disturbance component in the external folding disturbance component. Similarly, the system can also calculate the availability disturbance component, such as the decrease in success rate due to resource preemption, and the responsiveness disturbance component, such as the increase in latency due to queue congestion.

[0176] By deducting the corresponding external folding disturbance components from the local availability observation component, the local risk observation component, and the local responsiveness observation component, respectively, the local availability component, local risk component, and local responsiveness component of the corresponding candidate payment channel in the target area are obtained.

[0177] Based on the local availability component, the local risk component, and the local responsiveness component, the intrinsic availability status of the corresponding candidate payment channel in the target area is determined.

[0178] The three types of observed components currently observed in the target area are not entirely equivalent to the local true state of the candidate payment channel in the target area. Some changes may come from the target area's own recent call behavior, transaction input structure, and local link conditions. Other changes may be external influences transmitted from other areas around the same payment channel, such as high-pressure calls, high-risk inputs, continuous failures, or persistent biases, through shared risk control links, shared authorization queues, shared compensation links, or channel-side protection mechanisms. If these external influences are not separated from the observed components, and channel judgments are made directly based on the current observation results, the resulting scheduling conclusions are essentially still based on a mixed state. This can easily lead to misinterpreting short-term risk increases, response tailing, or acceptance contraction caused by other areas as the inherent state of the channel itself within the target area. Therefore, in this embodiment, the so-called "separate deduction" is not a formal subtraction of the three components, but rather a corresponding splitting of the parts of the three observed components that can be attributed to external regional effects according to the interference type, direction, intensity, and duration recorded in the folded interference edge. This restores the current observed state to a component expression that is closer to the local true state.

[0179] In this embodiment, for each candidate payment channel, all inter-regional folded interference edges pointing to the target region and corresponding to that candidate payment channel can be extracted from the regional state folding graph. These edges are then categorized according to their interference manifestation type, forming sets of available interference edges, risky interference edges, and responsive interference edges. Subsequently, the source region action feature summary, interference intensity level, effective time range, and timeliness marker carried in each interference edge are read, and it is determined whether the interference edge should still participate in the de-interference processing at the current scheduling moment. For interference edges that have exceeded their effective time range or have entered a decay state in the most recent update cycle, their participation level can be reduced or they can be excluded from this stripping process. For interference edges that are still in the current cycle's effective state and are consistent with the current observed change direction of the target region, they are retained as valid external disturbance sources. Specifically, in the separation of local availability observation components, interference edges related to continuous failures, reduced acceptance, channel degradation, and declining effective completion rates can be extracted first. Combined with the strength level and proximity of the effective time of these interference edges, external availability folding disturbance components can be identified. If the availability observation component of a candidate payment channel in the target region recently declines from a high availability range to a medium availability range, and there are strong interference edges of continuous failures from other regions in the graph, and the effective time of these interference edges covers the current observation window, then a portion of this decline can be interpreted as an impact of external acceptance reduction, and the corresponding disturbance portion can be deducted from the local availability observation component. For the separation of local risk observation components, interference edges related to high-risk clustering inputs, rule rejection propagation, and stricter risk review can be extracted. It can be observed whether these interference edges correspond to phenomena such as an increase in the current risk interception ratio, clustering of sensitive failure causes, and continuous occurrence of risk events in the target region. If the correspondence is valid, then the external risk folding disturbance component can be separated from the local risk observation component; that is, the currently observed increase in risk in the target region is not entirely considered local risk, but rather the portion that propagates synchronously with the high-risk effects in external regions is separated out. For local responsive observation components, interference edges related to authorized queue congestion, timeout propagation, compensation link accumulation, and retry dependency diffusion are extracted. Combined with recent observations of increased return latency, increased timeout events, and increased closed-loop ratio after retry in the target region, external responsive folding disturbance components are identified. To ensure that the stripping action does not exceed the current observation change itself, a local minimum constraint is added during the stripping process of each component. This means that after removing the external disturbance component, the minimum state residue sufficient to explain the recent behavior of the target region is retained, avoiding the removal of local anomalies. In this way, the local availability observation component, local risk observation component, and local responsive observation component are restored to the local availability component, local risk component, and local responsive component, respectively, all of which have had folding effects attributable to external region action paths removed.

[0180] It should be noted that the specific method for determining whether an interfering edge should still participate in the descrambling process at the current scheduling time can be determined based on the effective time, interference intensity, and direction of change. The effective time can be determined by superimposing the historical propagation delay on the time when the anomaly occurred in the source region. The historical propagation delay can be taken as the average interval from the occurrence of similar interference in the source region to the change of the observed components in the target region over the past 7 days, for example, 3 minutes. The validity period can be taken as two scheduling cycles, for example, 5 minutes per cycle, then the validity period is 10 minutes. If the current scheduling time falls within this validity period, and the interference intensity is in the top 50% of the intensity of similar interference over the past 30 days, and is consistent with the current observed change direction in the target region, then it will be included in this descrambling process.

[0181] Furthermore, after obtaining the local availability component, local risk component, and local responsiveness component, these three local components need to be reorganized into an intrinsic availability state that can be used for scheduling judgment of the current candidate payment channel. The intrinsic availability state in this application does not simply refer to a binary result of "available or unavailable," but rather a comprehensive expression of the local true acceptance value of the candidate payment channel in the target area after descrambling. Its formation process needs to consider the synergistic relationship between the three local components simultaneously. Specifically, the local availability component can first be interval-determined to determine whether the candidate payment channel still has a stable acceptance foundation in the target area; then, the local risk component can be hierarchically determined to determine whether the channel has entered a high-sensitivity review zone or is still in an acceptable risk zone in the target area; then, the local responsiveness component can be state-determined to determine whether the channel currently maintains a smooth return capability or has entered a tailing response or compensation dependency state. Based on this, the final intrinsic availability state is generated according to the preset intrinsic state combination rules. For example, when the local availability component is in a high availability range, the local risk component is in a normal risk range, and the local responsiveness component is in a stable response range, the intrinsic availability status of the candidate payment channel can be determined as a priority acceptance status. When the local availability component maintains a medium acceptance capacity, the local risk component has not entered the high-sensitivity review range, but the local responsiveness component has shown local tailing, it can be determined as a restricted acceptance status. When the local availability component has shrunk, the local risk component remains high, and the local responsiveness component simultaneously shows timeout accumulation, it can be determined as a suppressed acceptance status. The interval boundaries and combination rules here can be preset according to the historical normal distribution of the channel in the region. For example, the normal range of the local availability component, the normal range of the risk component, and the normal range of the responsiveness component of the payment channel in the target region in the past few scheduling cycles can be statistically analyzed first, and then the cases that deviate from the upper or lower bound of the normal range by more than a preset range can be classified into the shrinkage, high sensitivity, or tailing levels, respectively.

[0182] In the third aspect, regarding the method for determining the target payment channel, this embodiment adopts the following approach:

[0183] The intrinsic availability status score of each candidate payment channel is generated based on the intrinsic availability status.

[0184] Specifically, a preset scoring function can be used to map the local availability component, local risk component, and local responsiveness component in the intrinsic availability state into a numerical score.

[0185] The preset scoring function can be constructed using a combination of hierarchical mapping and weighted fusion. Specifically, it includes: determining the corresponding sub-scores based on the interval level, direction of change, and deviation from the historical normal range of the target area for the local availability component, local risk component, and local responsiveness component; and then, based on the business focus of the target area, the recent acceptance characteristics of the payment channel, and the influence of each sub-score on the channel's schedulability, weighted fusion is performed on the sub-scores to generate the intrinsic availability status score of the corresponding candidate payment channel.

[0186] It is understood that the calculation of specific sub-item scores can be implemented using existing segmented normalized scoring methods, and this application does not impose a unique limitation on its specific form. For example, each sub-item score can be uniformly mapped to 0 to 100 points. The higher the local availability component, the higher the corresponding availability sub-item score; the higher the local risk component, the lower the corresponding risk sub-item score; the higher the local responsiveness component, the higher the corresponding responsiveness sub-item score. The high, medium, and low intervals corresponding to each component can be determined based on the historical normal range of similar payment requests in the target area over the past 30 days. For example, if the 25th quartile of the local availability component over the past 30 days is 0.86, the 50th quartile is 0.92, and the 75th quartile is 0.96, then values ​​below 0.86 can be defined as the low availability interval, values ​​between 0.86 and 0.96 as the normal availability interval, and values ​​above 0.96 as the high availability interval. If the current local availability component is 0.94, then the availability sub-item score can be determined according to its relative position in the normal availability interval, for example, 80 points.

[0187] Furthermore, when weighting and integrating the sub-scores, basic weights can be pre-set based on the business focus of the target region, and then slightly adjusted based on recent channel performance. For example, for a target region that prioritizes payment success rate, the basic weights for availability, risk, and responsiveness sub-scores can be set to 0.50, 0.30, and 0.20, respectively. If the risk interception rate in this region has increased by more than 30% in the past 7 days compared to the average of the past 30 days, the risk weight is increased to 0.40, and the availability weight is correspondingly reduced to 0.40. Assuming a candidate payment channel has an availability score of 82, a risk score of 70, and a responsiveness score of 90, with current weights of 0.50, 0.30, and 0.20 respectively, then the intrinsic availability score is 80.

[0188] Based on the intrinsic availability score of each candidate payment channel, the candidate payment channel with the highest intrinsic availability score is determined as the target payment channel, and the pending payment request is scheduled to the target payment channel. Specifically, the system selects the channel with the highest score among all candidate channels as the target channel and performs routing scheduling operations. Since this score is generated based on intrinsic availability, it eliminates interference from scheduling behaviors in other areas. Therefore, the selected channel has optimal actual availability in the target area, thereby avoiding scheduling failures due to misjudgment of status and significantly improving the payment success rate.

[0189] In yet another example, this application embodiment will be described in detail using a specific cross-border payment scenario as an example.

[0190] The scenario is as follows: In a multi-region payment gateway system, there are two regions, A and B, which share the same underlying payment channel X. At a certain moment, region A experiences a surge in payment requests due to a large-scale promotional event.

[0191] First, the system executes the step of acquiring regional scheduling behavior information for each region. In this step, the system monitors that region A exhibits an "extremely high" intensity of call to payment channel X within a preset statistical time window, and the proportion of high-risk transactions in the transaction type characteristics has significantly increased. Meanwhile, the scheduling behavior information for region B remains stable. This behavioral information is captured and recorded in real time as a basis for subsequent analysis.

[0192] Secondly, the system executes the step of constructing a regional state folding graph based on regional scheduling behavior information and regional side feedback information. During the construction process, the system identified that the high-intensity, high-risk scheduling behavior of region A significantly increased the sensitivity of the underlying risk control model of payment channel X. This increase not only affected the transaction processing of region A itself, but also had a cross-regional impact on region B through the shared risk control link mechanism. Specifically, a folding interference edge was generated in the regional state folding graph, pointing from region A to region B, and the weight of this edge reflected the intensity of the increase in risk control sensitivity. This means that the scheduling behavior of region A changed the interpretative environment of the payment channel X state for region B, i.e., "state folding" occurred.

[0193] Subsequently, when region B initiates a pending payment request, the system, targeting region B, performs steps to determine candidate payment channels and strip the current observation state of folded perturbations. At this point, the current observation state of payment channel X from region B's perspective shows a significant decrease in its local availability, with an observation failure rate as high as 25%. Using traditional scheduling techniques, the system would directly determine channel X as unavailable based on this failure rate, thus eliminating or downgrading it, forcing requests from region B to be routed to less desirable channels with higher costs or longer paths. However, in this embodiment, the system identifies strong folding interference from region A to region B based on the region state folding graph. The system executes a folding perturbation stripping algorithm: extracting the external folding perturbation component caused by interference from region A from the observation failure rate. Calculations show that this external folding perturbation component is approximately 20%. The system performs a deduction operation, obtaining a failure rate of only 5% corresponding to the intrinsic availability of payment channel X in region B. This indicates that payment channel X actually still possesses extremely high real availability in region B; the previous unavailability was merely an "illusion" caused by interference from region A.

[0194] Finally, the system executes the step of determining the target payment channel based on the intrinsic availability status. Since the intrinsic availability status score after stripping shows that payment channel X has the optimal actual processing capacity in region B, the system ultimately schedules the pending payment request from region B to payment channel X. Actual processing results show that the request was successfully completed without triggering risk control interception.

[0195] This invention also discloses a multi-regional payment gateway intelligent routing scheduling system, comprising:

[0196] The regional information generation module is used to generate regional scheduling behavior information corresponding to each region and regional side feedback information of each payment channel in each region. The regional scheduling behavior information is used to characterize the call intensity, transaction type distribution, failure clustering and scheduling bias of each payment channel in the corresponding region. The regional side feedback information is used to characterize the local availability, risk and responsiveness of the payment channel in the corresponding region.

[0197] The folded graph construction module constructs a regional state folded graph based on the regional scheduling behavior information and the regional side feedback information. The nodes of the regional state folded graph are regional nodes, and the edges are inter-regional folding interference edges corresponding to the payment channels. The regional state folded graph is used to characterize the folding impact of the scheduling behavior of different regions on the state manifestation of each payment channel in different regions.

[0198] The candidate channel analysis module is used to determine the target area corresponding to the pending payment request when there is a pending payment request, determine multiple candidate payment channels corresponding to the target area based on the area state folding map, and perform folding perturbation stripping on the current observation state of each candidate payment channel in the target area to obtain the intrinsic available state corresponding to each candidate payment channel.

[0199] The routing and scheduling module is used to determine the target payment channel corresponding to the pending payment request based on the intrinsic availability status of each candidate payment channel, and to schedule the pending payment request to the target payment channel.

[0200] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A multi-region payment gateway intelligent routing dispatching method, characterized in that, Each region corresponds to multiple payment channels, and each payment channel corresponds to different regional feedback information in different regions. The regional feedback information is used to characterize the local availability, risk, and responsiveness of the payment channel in the corresponding region. The method includes: When there are pending payment requests, obtain the corresponding regional scheduling behavior information for each region; Based on regional scheduling behavior information and regional side feedback information, a regional state folding diagram is constructed; For the target area corresponding to the payment request to be processed, based on the area state folding map, multiple candidate payment channels are determined, and the current observation state of the candidate payment channels in the target area is folded and perturbed to obtain the intrinsic available state corresponding to each candidate payment channel. Based on the intrinsic availability status, the target payment channel corresponding to the pending payment request is determined, and the pending payment request is scheduled to the target payment channel.

2. The intelligent routing and scheduling method for multi-regional payment gateways according to claim 1, characterized in that, The method for determining the regional side feedback information includes: Obtain historical payment processing data for historical payment requests in different regions and payment channels. The historical payment processing data includes at least the region identifier, payment channel identifier, processing result identifier, failure reason identifier, result return delay information, and retry processing information. The historical payment processing data is classified and aggregated according to the region identifier and payment channel identifier to obtain the set of historical feedback events for each payment channel in each region. Based on the distribution of successful processing events, failed processing events, risk interception events, timeout events, and retry events in each historical feedback event set, the local availability, local risk, and local responsiveness of the corresponding payment channel in the corresponding region are determined. Based on the local availability, local risk, and local responsiveness, regional feedback information for the corresponding payment channel in the corresponding region is generated.

3. The intelligent routing and scheduling method for multi-regional payment gateways according to claim 1, characterized in that, The regional scheduling behavior information is used to characterize the call intensity, transaction type distribution, failure clustering, and scheduling bias of each payment channel within the corresponding region. This regional scheduling behavior information is obtained through the following methods: Obtain the scheduling results of each region for each payment channel within the preset statistical time window; The scheduling result information is aggregated by region based on the region identifier, and the previous scheduling result corresponding to each payment channel is determined in each region according to the payment channel identifier. Based on the previous scheduling results for each payment channel, generate regional scheduling behavior information for the corresponding area.

4. The intelligent routing and scheduling method for multi-regional payment gateways according to claim 3, characterized in that, If a region does not have a previous scheduling result for a payment channel within a preset statistical time window, the region's scheduling behavior information is generated in the following ways: Based on the regional state folding diagram of the previous scheduling cycle, a reference region corresponding to the region is determined, and the previous scheduling result of the corresponding payment channel in the reference region is extracted. The previous scheduling result is corrected by regional folding to generate a substitute scheduling result for the corresponding payment channel in the reference area, and corresponding regional scheduling behavior information is generated based on the substitute scheduling result.

5. The intelligent routing and scheduling method for multi-regional payment gateways according to claim 3, characterized in that, The method for constructing the region state folding map includes: Based on the regional scheduling behavior information corresponding to each region, the scheduling effect characteristics of each region on each payment channel within the current scheduling cycle are determined. The scheduling effect characteristics include at least the call intensity effect characteristics, transaction type effect characteristics, failure clustering effect characteristics, and scheduling bias effect characteristics. Based on the regional feedback information of each payment channel in each region, the status display characteristics of each payment channel in each region are determined. The status display characteristics include at least local availability display characteristics, local risk display characteristics, and local responsiveness display characteristics. Based on the correlation between the scheduling characteristics of each region and the status display characteristics of each payment channel in each region, the cross-regional folding interference relationship between the scheduling behavior of different regions and the status display of each payment channel in other regions is determined. Based on the cross-regional folding interference relationship between regions, a regional state folding diagram is constructed. The regional state folding diagram is used to characterize the folding effect of the scheduling behavior of different regions on the state manifestation of each payment channel in different regions. The nodes of the regional state folding diagram are regional nodes, and the edges are the inter-regional folding interference edges corresponding to the payment channels.

6. The intelligent routing and scheduling method for multi-regional payment gateways according to claim 1, characterized in that, The method for determining the target region includes: Obtain the region identification information corresponding to the payment request to be processed. The region identification information includes at least one of the following: payment initiation region identifier, payment terminal home region identifier, account home region identifier, currency identifier, and merchant access region identifier. Based on the region identification information, the region affiliation result corresponding to the payment request to be processed is determined; The region corresponding to the region attribution result is determined as the target region.

7. The intelligent routing and scheduling method for multi-regional payment gateways according to claim 5, characterized in that, The method for determining the candidate payment channel includes: Based on the target area, obtain multiple payment channels within the area corresponding to the target area, as an initial payment channel set; Obtain the current observation status of each payment channel in the initial payment channel set in the target area, and based on the area state folding map, determine the folding interference of other areas on each payment channel in the target area; Based on the current observation status of each payment channel and the folding interference situation, determine the channel suppression status of each payment channel; Payment channels whose channel suppression status meets the preset suppression conditions are eliminated, and the remaining payment channels are identified as the multiple candidate payment channels.

8. The intelligent routing and scheduling method for multi-regional payment gateways according to claim 7, characterized in that, The folding disturbance stripping includes: Extract the local availability observation component, local risk observation component, and local responsiveness observation component from the current observation status of each candidate payment channel in the target area; Extract the inter-regional folding interference edges from other regions pointing to the target region and corresponding to the corresponding candidate payment channel from the regional state folding diagram, and determine the external folding disturbance component of the corresponding candidate payment channel in the target region based on the folding interference intensity corresponding to each inter-regional folding interference edge. By deducting the corresponding external folding disturbance components from the local availability observation component, the local risk observation component, and the local responsiveness observation component, respectively, the local availability component, local risk component, and local responsiveness component of the corresponding candidate payment channel in the target area are obtained; Based on the local availability component, the local risk component, and the local responsiveness component, the intrinsic availability status of the corresponding candidate payment channel in the target area is determined.

9. The intelligent routing and scheduling method for multi-regional payment gateways according to claim 1, characterized in that, The method for determining the target payment channel includes: The intrinsic availability status score of each candidate payment channel is generated based on the intrinsic availability status. Based on the intrinsic availability score of each candidate payment channel, the candidate payment channel with the highest intrinsic availability score is determined as the target payment channel, and the pending payment request is scheduled to the target payment channel.

10. A multi-regional payment gateway intelligent routing scheduling system, used to implement the multi-regional payment gateway intelligent routing scheduling method as described in any one of claims 1-9, characterized in that, The system includes: The regional information generation module is used to generate regional scheduling behavior information corresponding to each region and regional side feedback information of each payment channel in each region. The regional scheduling behavior information is used to characterize the call intensity, transaction type distribution, failure clustering and scheduling bias of each payment channel in the corresponding region. The regional side feedback information is used to characterize the local availability, risk and responsiveness of the payment channel in the corresponding region. The folded graph construction module constructs a regional state folded graph based on the regional scheduling behavior information and the regional side feedback information. The nodes of the regional state folded graph are regional nodes, and the edges are inter-regional folding interference edges corresponding to the payment channels. The regional state folded graph is used to characterize the folding impact of the scheduling behavior of different regions on the state manifestation of each payment channel in different regions. The candidate channel analysis module is used to determine the target area corresponding to the pending payment request when there is a pending payment request, determine multiple candidate payment channels corresponding to the target area based on the area state folding map, and perform folding perturbation stripping on the current observation state of each candidate payment channel in the target area to obtain the intrinsic available state corresponding to each candidate payment channel. The routing and scheduling module is used to determine the target payment channel corresponding to the pending payment request based on the intrinsic availability status of each candidate payment channel, and to schedule the pending payment request to the target payment channel.

Images