Supplier issuing payment method based on intelligent batching and elastic routing
Through the supplier payment method of intelligent batch and elastic routing, the batch parameters and payment paths are dynamically adjusted, and the problems of low capital utilization rate and poor abnormal recovery ability in the existing technology are solved, and efficient and stable payment processing is achieved.
Patent Information
- Application Number
- CN202510357075.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-25
- Publication Date
- 2025-07-11
AI Technical Summary
The existing supplier payment system cannot be dynamically optimized when paying tasks in batches, the capital utilization rate is low, the abnormal recovery ability is poor, and it is difficult for multiple suppliers to achieve intelligent routing decisions.
The intelligent batch model and elastic routing mechanism are adopted, combined with deep reinforcement learning and graph attention network, and the batch parameters and payment paths are dynamically adjusted to achieve self-healing exception handling.
The capital utilization rate has been improved to more than 90%, the task completion time has been shortened to the minute level, the manual intervention rate and potential costs have been reduced, and the stability and efficiency of the payment business have been ensured.
Smart Images
Figure CN120297968A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of Internet crowdsourcing platforms, and particularly relates to a method for issuing payments to suppliers based on intelligent batching and elastic routing. Background Art
[0002] When the existing supplier payment system adopted by the crowdsourcing platform issues payment tasks in batches, it uses static batching rules but does not solve the problem of dynamic optimization; some systems adopt multi-level supplier collaboration, but the routing decision depends on manual configuration. Therefore, there is an urgent need for an intelligent and highly available supplier payment method.
[0003] The existing supplier payment methods have the following technical defects:
[0004] 1. Rigid batching rules: Using a fixed amount ceiling (such as 500,000 yuan per batch), it cannot be dynamically adjusted according to the real-time load, and the capital utilization rate is less than 60%;
[0005] 2. Poor exception recovery ability: Manual intervention is required when the payment fails, and the average recovery time exceeds 4 hours;
[0006] 3. Difficulty in multi-supplier collaboration: Lack of an intelligent routing mechanism, and it cannot automatically switch to an alternative path when the interface fails. Summary of the Invention
[0007] The purpose of the present invention is to provide a method for issuing payments to suppliers based on intelligent batching and elastic routing, which is used to solve the technical problems in the prior art that cannot be dynamically optimized and has a low capital utilization rate in the aspect of issuing payment tasks in batches.
[0008] The method for issuing payments to suppliers based on intelligent batching and elastic routing includes:
[0009] A. Input the collected data into the intelligent batching model to output the batching method, and then generate payment orders in sequence;
[0010] B. Based on the directed graph established by the supplier network, use the graph attention network to calculate the optimal path of the payment path and determine the payment path;
[0011] C. Process the abnormal situations in the process of issuing payments based on the self-healing exception handling mechanism.
[0012] Preferably, the intelligent batching model adopts a deep reinforcement learning model, and the reward function of the intelligent batching model is R = λ1S s +λ2T p +λ3min(1,BthresholdBcurrent), where S s represents the payment success rate, T pFor the processing time consumption, Bthreshold and Bcurrent are the fund risk parameters, λ1, λ2, and λ3 are the weights of the payment success rate, processing time consumption, and fund risk in sequence, and the sum of the three is 1, and λ1 is the largest; the intelligent batch model uses an algorithm based on policy gradient to learn how to select the optimal action according to the state, and uses the Q function to evaluate the long-term cumulative reward of taking a certain action in a certain state; after each decision of the model, it continuously adjusts its own policy according to the actually obtained reward value to seek the best batch parameter combination.
[0013] Preferably, the state space expression of the intelligent batch model is:
[0014]
[0015] Among them, S t represents the state space, b k represents the account balance information of the k-th supplier, c k represents the interface response delay data of the k-th supplier, d k represents the historical payment success rate of the k-th supplier, and e represents the real-time load condition of the system. The value of the real-time load condition is calculated by comprehensively considering the number of other payment tasks in progress in the system, the CPU and memory occupancy rates of system resources, etc.
[0016] Preferably, the action space of the intelligent batch model is the output dynamic batch parameters, and the expression is:
[0017]
[0018] Among them, α i represents the i-th amount of the batch payment, and β i represents the time interval from the i-th payment to the next payment in the batch payment.
[0019] Preferably, in the action space, the single-limit of the i-th amount is flexibly adjusted within a certain range, and the interval time from the i-th amount to the next payment varies within a certain range; when performing intelligent dynamic batch processing, time periods are divided, including low-load periods and high-load periods. The low-load period is the time period with fewer payment tasks. At this time, the single-limit of the i-th amount is increased, and the upper limit of the interval time is decreased; the high-load period is the peak business period. At this time, the single-limit of the i-th amount is decreased, and at the same time, the upper limit of the interval time is increased.
[0020] Preferably, Bcurrent represents the current balance of the supplier's account, and Bthreshold is a pre-set risk threshold. When Bcurrent approaches or exceeds Bthreshold, it indicates that the supplier's capital risk is low. When Bcurrent approaches or exceeds Bthreshold, it indicates that the supplier's capital risk is low. The degree to which Bcurrent exceeds Bthreshold is represented by a capital risk parameter, and the reward function measures the capital risk through min(1, Bthreshold / Bcurrent).
[0021] Preferably, the algorithm based on policy gradient is the Proximal Policy Optimization algorithm. When performing intelligent dynamic batch processing, policy updates are performed regularly based on time.
[0022] Preferably, in step B, the graph attention network calculates the attention of the nodes in the directed graph. According to the weight information of the nodes and edges, it accurately evaluates the advantages and disadvantages of different paths from the starting node to the target node. Among them, the starting node is the payment initiation platform, and the target node is the final payee; the supplier network is abstracted as a directed graph G=(V, E), where the node V represents the nodes in the payment path, including each supplier and the payment initiation platform, and the supplier includes the final payee in the payment path. The edge E represents the connection relationship between suppliers.
[0023] Preferably, the edge weight W ij in the directed graph is calculated by the formula: W ij =1 / (D ij ×R ij ×F ij ), where W ij represents the edge weight from node i to node j, D ij represents the interface delay from node i to node j, R ij represents the rate from node i to node j, and F ij represents the payment failure rate from node i to node j; perform the optimal path calculation, and determine the payment path according to the sum of the weights of each edge of the path.
[0024] Preferably, step C classifies and processes the exceptions that occur during the payment process, specifically including L1 exceptions, L2 exceptions, and L3 exceptions;
[0025] The L1 exception is a single failure, and the handling method is: when the system detects a single payment failure, immediately start the automatic retry mechanism;
[0026] The L2 exception is insufficient balance, and the handling method is: if the supplier's account balance is insufficient, the system disassembles the current batch payment amount;
[0027] The L3 exception is an interface timeout, and the handling method is: once the interface timeout exception is detected, the system quickly switches to the backup supplier within a short period of time.
[0028] Preferably, in the handling method of the L1 exception, each time a retry is made, the system will wait briefly for a period of time and then initiate the payment request again to cope with possible payment failures caused by temporary reasons such as network fluctuations. There is an upper limit to the number of automatic retries.
[0029] In the handling method of the L2 exception, the payment amount of the current batch is automatically split into multiple small payments by the system.
[0030] In the handling method of the L3 exception, through the pre-constructed backup supplier list and fast switching mechanism, an available supplier interface can be reselected within an extremely short period of time to continue the payment operation.
[0031] The advantages of the present invention are as follows:
[0032] 1. Efficiency improvement: Through intelligent dynamic batch processing, the peak payment throughput of the present invention can reach 2,000 transactions per minute. The dynamically adjusted batch parameters can reasonably arrange the payment rhythm according to the real-time load of the system and the supplier status, avoid interface congestion, and greatly improve the payment processing speed. At the same time, the elastic routing mechanism ensures that the payment path always remains efficient, reducing delays caused by path problems. The task completion time is greatly shortened from the traditional hourly level to the minute level, greatly improving the capital turnover efficiency and meeting the timeliness requirements for large-scale and high-frequency payments in scenarios such as flexible employment and crowdsourcing services.
[0033] 2. Risk controllability: The real-time monitoring mechanism of the present invention works closely with the self-healing exception handling mechanism, and can detect and solve more than 85% of payment exception events in a timely manner. For example, for common exceptions such as single payment failure, insufficient balance, and interface timeout, the system can automatically perform operations such as retry, split batches, or switch suppliers without manual intervention. The manual intervention rate is reduced by 90%, effectively reducing the risks brought by human errors. At the same time, the exception recovery time is greatly shortened, ensuring the stability of the payment business and the security of funds.
[0034] 3. Cost optimization: Based on the dynamic batch strategy of deep reinforcement learning, the present invention can optimize the capital allocation according to real-time data, and the capital utilization rate is increased to more than 90%, making the capital more efficiently utilized and reducing the capital cost. The elastic routing mechanism analyzes the supplier network through graph neural networks and selects the optimal path, reducing the payment delay by 40% and reducing the potential cost losses caused by delays. At the same time, by reasonably selecting suppliers with lower rates, the supplier rate cost is effectively reduced by 15%, realizing the optimization of the overall payment cost. Brief Description of the Drawings
[0035] Figure 1 This is the basic flowchart of a payment method for suppliers based on intelligent batch splitting and elastic routing in the present invention. Detailed implementation manners
[0036] The following is a more detailed description of the specific implementation manners of the present invention by referring to the accompanying drawings and describing the embodiments, so as to help those skilled in the art have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.
[0037] As Figure 1 shown, the present invention provides a payment method for suppliers based on intelligent batch splitting and elastic routing, including the following contents.
[0038] A. Input the collected data into the intelligent batch splitting model to output the batch splitting method, and then generate payment orders in sequence.
[0039] The intelligent batch splitting model adopts a deep reinforcement learning (DRL) model. The state space expression of the intelligent batch splitting model is:
[0040]
[0041] Among them, S t represents the state space, b k represents the account balance information of the kth supplier, c k represents the interface response delay data of the kth supplier, d k represents the historical payment success rate of the kth supplier, and e represents the real-time load condition of the system. The value of the real-time load condition is calculated by comprehensively considering aspects such as the number of other payment tasks in progress in the system, the CPU and memory occupancy rates of the system resources, etc.
[0042] The action space of the intelligent batch splitting model is the output dynamic batch parameters, and the expression is:
[0043]
[0044] Among them, α i represents the i-th amount of batch payment, β i represents the time interval from the i-th payment to the next payment in batch payment. In the action space, the single-limit of the i-th amount is flexibly adjusted within a certain range (300,000 - 1,000,000), and the interval time from the i-th amount to the next payment varies within a certain range (1 - 10 minutes).
[0045] The reward function of the intelligent batch splitting model is R = λ1S s + λ2T p+λ3min(1, Bthreshold / Bcurrent), where the reward function is the key to guiding the model to optimize the strategy; among them, S s (Ssuccess) represents the payment success rate. A higher success rate means a smooth payment process and less impact on the overall business, so it has a higher proportion in the reward function; T p (Tprocess1) is the processing time consumption. The shorter the payment processing time, the higher the capital turnover efficiency, which is reflected in the form of Tprocess1; Bthreshold / Bcurrent is the capital risk parameter. Bcurrent represents the current balance of the supplier's account, and Bthreshold is the pre-set risk threshold. When Bcurrent is close to or exceeds Bthreshold, it indicates that the supplier's capital risk is low. The degree to which Bcurrent exceeds Bthreshold is represented by the capital risk parameter, and the reward function measures the capital risk through min(1, Bthreshold / Bcurrent); λ1, λ2, and λ3 are the weights of the payment success rate, processing time consumption, and capital risk in turn. The sum of the three is 1, and λ1 is the largest. In this embodiment, λ1, λ2, and λ3 are 0.4, 0.3, and 0.3 in turn.
[0046] The intelligent batch model uses an algorithm based on policy gradients to learn how to select the optimal action according to the state. The algorithm used can be the Proximal Policy Optimization algorithm (PPO), and the Q function is used to evaluate the long-term cumulative reward for taking a certain action in a certain state. After each decision, the model continuously adjusts its own strategy according to the actually obtained reward value to seek the best combination of batch parameters.
[0047] By constructing a multi-layer neural network structure, the intelligent batch model can perform complex feature extraction and pattern recognition on the input data. It receives data from multiple data sources. After being trained and optimized with a large amount of data, the model can accurately output appropriate dynamic batch parameters according to these input data.
[0048] When performing intelligent dynamic batch processing, this step updates the policy regularly based on time to adapt to the changing business environment. In the embodiment, the model updates the policy every 5 minutes. At the same time, this step also divides time periods, including low-load periods and high-load periods. The low-load period is the time period with fewer payment tasks, such as 1:00 - 5:00 in the early morning. At this time, the system resources are sufficient, the network pressure is small, and the interface response speed is fast. Therefore, this step increases the single-limit of the i-th amount to a larger value in the corresponding interval, for example, adjusts the single-limit from 500,000 to 800,000, and at the same time appropriately shortens the payment interval, that is, reduces the upper limit of the interval of the interval time, which will speed up the payment process and improve the efficiency of fund disbursement; while in the high-load period, due to the high system load and increased interface response delay during the business peak period, it is necessary to reduce the single-limit of the i-th amount, for example, adjust the single-limit from 500,000 to 300,000, and at the same time increase the upper limit of the interval of the interval time, which can extend the interval time to a certain extent, avoid interface congestion or even collapse caused by a large number of requests in a short time, and ensure the stability of the payment process.
[0049] B. Based on the directed graph established by the supplier network, use the graph attention network to implement the optimal path calculation of the payment path, thereby determining the payment path.
[0050] While performing intelligent dynamic batch processing, this step uses the graph attention network (GAT) to implement the optimal path calculation of the payment path. The GAT model can perform attention calculation on the nodes in the directed graph, and accurately evaluate the pros and cons of different paths from the starting node to the target node according to the weight information of the nodes and edges. Among them, the starting node is the payment initiation platform, and the target node is the final recipient.
[0051] Before establishing the GAT model, abstract the supplier network as a directed graph G=(V, E), where the node V represents the nodes in the payment path, including each supplier and the payment initiation platform. The supplier includes the final recipient of the payment path, and the edge E represents the connection relationship between the suppliers. In the directed graph, a specific weight is assigned to each node, and this weight comprehensively considers important factors such as supplier interface delay, rate, and payment failure rate; the interface delay directly affects the timeliness of payment, the rate is related to the payment cost, and the payment failure rate reflects the reliability of the supplier. The edge weight W ij The calculation formula is: W ij =1 / (D ij ×R ij ×F ij ), W ij represents the edge weight from node i to node j, D ij represents the interface delay from node i to node j, R ij represents the rate from node i to node j, F ijIt represents the payment failure rate from node i to node j. The shorter the interface delay, the lower the rate, and the lower the failure rate, the better the connection relationship represented by this edge, that is, the greater the edge weight. In this way, the complex relationships of the entire supplier network are clearly presented in the form of a mathematical model, providing a basis for subsequent path decisions.
[0052] For example, if the payment path includes a primary path and a backup path, where the primary path is "Cloud Gongxiang Platform → Supplier A → Service Provider X → Individual", and the backup path is "Cloud Gongxiang Platform → Supplier B → Service Provider Y → Individual", where the "Cloud Gongxiang Platform" is the payment initiation platform, the "Individual" is the ultimate payee who actually provides the service, and "Supplier A", "Supplier B", "Service Provider X", and "Service Provider Y" are all suppliers in the payment path. When calculating the optimal path, if the sum of the edge weights of the primary path is 0.8 and the sum of the edge weights of the backup path is 0.7, it means that a certain failure occurs in the nodes on the primary path (such as the interface timeout of Supplier A), resulting in the sum of the edge weights of the primary path exceeding the sum of the edge weights of the backup path. At this time, according to the output of the GAT model, the system automatically switches the payment path from the primary path to the backup path to ensure the continuity of the payment service.
[0053] C. Handle the exceptions during the issued payment process based on the self-healing exception handling mechanism.
[0054] This step classifies and handles the exceptions that occur during the issued payment process, specifically as follows:
[0055] 1. L1 exception (single failure): When the system detects a single payment failure, it immediately starts the automatic retry mechanism. To avoid resource waste and business delays caused by infinite retries, the maximum number of automatic retries is set to 3 times. Each time a retry is made, the system will wait briefly for a period of time (such as 1 - 5 seconds, adjusted according to the actual situation), and then resend the payment request to handle possible payment failures caused by temporary reasons such as network fluctuations.
[0056] 2. L2 exception (insufficient balance): If the situation of insufficient balance in the supplier's account occurs, the system disassembles the current batch payment amount. For example, if the originally planned payment is 1 million and the supplier's balance is insufficient, the system automatically disassembles it into multiple small payments, such as 500,000 + 500,000, and attempts to pay in batches to increase the possibility of successful payment.
[0057] 3. L3 exception (interface timeout): Once an interface timeout exception is detected, the system quickly switches to a backup supplier within 1 minute. Through the pre-built backup supplier list and fast switching mechanism, it can re-select an available supplier interface within a very short time and continue the payment operation, minimizing the impact of the exception on the payment service.
[0058] The present invention has been described above in an exemplary manner with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited by the above methods. As long as various non-substantive improvements are made by adopting the inventive concept and technical solution of the present invention, or the inventive concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.
Claims
1. A supplier payment method based on intelligent batch and flexible routing, characterized in that: Including: A. Input the collected data into the intelligent batch model to output the batch mode, and then generate payment orders in sequence; B. Based on the directed graph established by the supplier network, use the graph attention network to calculate the optimal payment path and determine the payment path; C. Process the abnormal situations in the issued payment process based on the self-healing exception handling mechanism.
2. The payment method for suppliers issued based on intelligent batch and elastic routing according to claim 1, wherein: The intelligent batch model uses a deep reinforcement learning model, and the reward function of the intelligent batch model is R = λ1S s + λ2T p + λ3min(1, BthresholdBcurrent), where S s represents the payment success rate, T p is the processing time, BthresholdBcurrent is the capital risk parameter, λ1, λ2, and λ3 are the weights of the payment success rate, processing time, and capital risk in turn, and the sum of the three is 1, and λ1 is the largest; the intelligent batch model uses an algorithm based on policy gradient to learn how to select the optimal action according to the state, and uses the Q function to evaluate the long-term cumulative reward of taking a certain action in a certain state; after each decision, the model continuously adjusts its own policy according to the actual obtained reward value to seek the best batch parameter combination.
3. The supplier payment method based on intelligent batch and flexible routing according to claim 2, characterized in that: The state space expression of the intelligent batch model is: Among them, S t represents the state space, b k represents the account balance information of the k-th supplier, c k represents the interface response delay data of the k-th supplier, d k represents the historical payment success rate of the k-th supplier, and e represents the real-time load condition of the system. The value of the real-time load condition is calculated by comprehensively considering aspects such as the number of other payment tasks in progress in the system, the CPU and memory occupancy rates of system resources, etc.
4. The method for a supplier to issue payments based on intelligent batch and elastic routing according to claim 3, wherein: The action space of the intelligent batch model is the output dynamic batch parameter, and the expression is: where α i represents the i-th amount paid in batches, and β i represents the time interval from the i-th payment in batches to the next payment.
5. The supplier payment method based on intelligent batch and elastic routing according to claim 3 is characterized in that: In the action space, the single-limit of the i-th amount is flexibly adjusted within a certain range, and the interval time from the i-th amount to the next payment varies within a certain range; when performing intelligent dynamic batch processing, time periods are divided, including low-load periods and high-load periods. The low-load period is the time period with fewer payment tasks. At this time, the single-limit of the i-th amount is increased, and the upper limit of the interval time is decreased; The high-load period is the business peak period. At this time, the single-limit of the i-th amount is decreased, and at the same time, the upper limit of the interval time is increased.
6. The supplier payment method based on intelligent batch and elastic routing according to claim 2, wherein: Bcurrent represents the current balance of the supplier's account, and Bthreshold is the pre-set risk threshold. When Bcurrent approaches or exceeds Bthreshold, it indicates that the supplier's capital risk is low. When Bcurrent approaches or exceeds Bthreshold, it indicates that the supplier's capital risk is low. The degree to which Bcurrent exceeds Bthreshold is represented by the capital risk parameter, and the reward function measures the capital risk through min(1, Bthreshold / Bcurrent).
7. A method for a supplier to issue payments based on intelligent batch and elastic routing according to claim 2, characterized in that: The algorithm based on policy gradient is the proximal policy optimization algorithm. When performing intelligent dynamic batch processing, policy updates are made regularly based on time.
8. The supplier payment method based on intelligent batch and elastic routing according to claim 1, characterized in that: In step B, the graph attention network calculates the attention of the nodes in the directed graph. According to the weight information of the nodes and edges, it accurately evaluates the advantages and disadvantages of different paths from the starting node to the target node. Among them, the starting node is the payment initiation platform, and the target node is the final payee; the supplier network is abstracted as a directed graph G=(V, E), where the node V represents the nodes in the payment path, including each supplier and the payment initiation platform. The supplier includes the final payee in the payment path, and the edge E represents the connection relationship between suppliers.
9. The method for a supplier to issue payments based on intelligent batching and elastic routing according to claim 7, wherein: Edge weight W in a directed graph ij The calculation formula is: W ij = 1 / (D ij × R ij × F ij ), where W ij represents the edge weight from node i to node j, D ij represents the interface delay from node i to node j, R ij represents the rate from node i to node j, and F ij represents the payment failure rate from node i to node j; perform the optimal path calculation and determine the payment path according to the sum of the edge weights of each path.
10. A supplier payment method based on intelligent batch and elastic routing according to claim 1, characterized in that: Step C classifies and processes the abnormalities that occur in the issued payment process, specifically including L1 abnormalities, L2 abnormalities, and L3 abnormalities; The L1 abnormality is a single failure, and the processing method is: when the system detects a single payment failure, it immediately starts the automatic retry mechanism; The L2 abnormality is insufficient balance, and the processing method is: if the situation of insufficient balance in the supplier's account occurs, the system disassembles the current batch payment amount; The L3 abnormality is interface timeout, and the processing method is: once the interface timeout abnormality is detected, the system quickly switches to the backup supplier within a short time.
11. A supplier payment method based on intelligent batch and elastic routing according to claim 9, characterized in that: In the handling method of L1 exception, during each retry, the system will wait briefly for a period of time and then initiate a payment request again to handle possible payment failures caused by temporary reasons such as network fluctuations. There is an upper limit to the number of automatic retries. In the handling method of L2 exception, the payment amount of the current batch is automatically split into multiple small payments by the system. In the handling method of L3 exception, through a pre-built list of alternative suppliers and a fast switching mechanism, it is possible to re-select an available supplier interface within an extremely short time and continue with the payment operation.