Payment routing method based on deep learning and related equipment

Abstract

The invention provides a payment routing method based on deep learning and related equipment, and is suitable for the field of financial science and technology. Comprising the following steps: obtaining a target payment request and channel characteristics of a plurality of alternative payment channels; predicting a payment success probability and a risk control triggering probability of each alternative payment channel based on the transaction characteristics and the channel characteristics of the target payment request through a pre-trained payment prediction model; determining a plurality of candidate routing schemes through a preset multi-objective optimization algorithm by taking the payment success probability, the risk control triggering probability and the preset service charge rate of each alternative payment channel as optimization objectives; determining a target routing scheme from the plurality of candidate routing schemes according to a preset service strategy; and processing the target payment request based on the target routing scheme. According to the method and the device, the personalized routing decision can be dynamically generated for the payment request based on the real-time state of the payment channel and the characteristics of the user and the merchant, so that the success rate of payment transaction and the routing selection efficiency are improved.

Description

Technical Field

[0001] This application applies to the fields of fintech and artificial intelligence, and particularly relates to a payment routing method and related equipment based on deep learning. Background Technology In e-commerce and online services, the payment routing system is the core hub connecting merchants, users, and multiple third-party payment channels. Its main task is to select the optimal execution path for each payment request to ensure payment success rate, reduce transaction costs, and control risks.

[0002] In related technologies, traditional payment routing systems mostly adopt strategies based on static rules. For example, fixed channel priorities are preset for different businesses, and when the preferred channel fails, backup channels are tried in a predetermined order. Although such systems are simple to implement, as the scale of business expands and payment scenarios become more complex, their rigid routing strategies and lack of dynamic awareness of the real-time status of channels make it difficult to achieve personalized routing for different users and merchants, resulting in a low overall transaction success rate. Summary of the Invention

[0003] The main objective of this application is to propose a payment routing method and related equipment based on deep learning, which can dynamically generate personalized routing decisions for payment requests based on the real-time status of the payment channel and the characteristics of users and merchants, thereby improving the success rate of payment transactions and the efficiency of routing selection.

[0004] To achieve the above objectives, a first aspect of this application proposes a payment routing method based on deep learning, the method comprising: Obtain the target payment request and the channel characteristics of multiple alternative payment channels; Using a pre-trained payment prediction model, based on the transaction characteristics of the target payment request and the channel characteristics, the success probability and risk control trigger probability of each of the alternative payment channels are predicted. The payment success probability, the risk control trigger probability, and the preset handling fee rate of each of the alternative payment channels are used as optimization objectives. Multiple candidate routing schemes are determined through a preset multi-objective optimization algorithm. Based on a preset business strategy, a target routing scheme is determined from the multiple candidate routing schemes; The target payment request is processed based on the target routing scheme.

[0005] In some embodiments, the acquisition of the target payment request and the channel characteristics of multiple alternative payment channels includes: Extract transaction features, user features, and merchant features from the target payment request; Obtain the historical and real-time channel characteristics of each of the candidate payment channels; The user's geographic location in the user characteristics and the merchant's industry in the merchant characteristics are identified as category features; The transaction amount in the transaction features, the historical success rate in the historical channel features, and the network latency in the real-time channel features are normalized and used as numerical features.

[0006] In some embodiments, the payment prediction model includes a payment success probability prediction sub-model and a risk control trigger probability prediction sub-model. The pre-trained payment prediction model, based on the transaction characteristics of the target payment request and the channel characteristics, predicts the payment success probability and risk control trigger probability of each of the candidate payment channels, including: The categorical features and the numerical features are combined into model input features; The input features of the model are input into the payment success probability prediction sub-model to generate the payment success probability; wherein, the pre-training process of the payment success probability prediction sub-model includes: acquiring historical payment transaction data containing actual payment success results; constructing a first training input based on the historical payment transaction data, and constructing the actual payment success results as a first training label; and, based on the first training input and the second training label, calculating a first prediction loss value and iteratively adjusting the model parameters of the payment success probability prediction sub-model until the first prediction loss value is lower than a preset first loss value threshold. The model input features are input into the risk control trigger probability prediction sub-model to generate the risk control trigger probability; wherein, the pre-training process of the risk control trigger probability prediction sub-model includes: obtaining risk control interception logs containing actual risk control trigger results; constructing a second training input based on historical payment transaction data, and constructing the actual risk control trigger results as a second training label; and, based on the second training input and the second training label, calculating a second prediction loss value and iteratively adjusting the model parameters of the risk control trigger probability prediction sub-model until the second prediction loss value is lower than a preset second loss value threshold.

[0007] In some embodiments, the steps involve using the payment success probability, the risk control trigger probability, and the preset transaction fee rate of each of the alternative payment channels as optimization objectives, and determining multiple candidate routing schemes through a preset multi-objective optimization algorithm, including: Maximizing the payment success probability, minimizing the risk control trigger probability, and minimizing the transaction fee rate are taken as multiple optimization objectives; By comparing the multiple alternative payment channels pairwise, multiple candidate routing schemes are screened and determined. Each candidate routing scheme satisfies the following condition: no other alternative payment channel performs better than the candidate routing scheme in at least one of the optimization objectives, and performs no worse than the candidate routing scheme in all other optimization objectives.

[0008] In some embodiments, the preset service strategy includes: a cost-first strategy, a success-rate-first strategy, and a balanced strategy. The step of determining the target routing scheme from the plurality of candidate routing schemes according to the preset service strategy includes: When the preset business strategy is a cost-first strategy, the candidate routing scheme with the lowest transaction fee rate is selected from the multiple candidate routing schemes as the target routing scheme; When the preset business strategy is a success rate priority strategy, the candidate routing scheme with the highest payment success probability is selected from the multiple candidate routing schemes as the target routing scheme; When the preset service strategy is a balanced strategy, a comprehensive score is calculated for each candidate routing scheme according to the preset weight parameters, and the candidate routing scheme with the highest comprehensive score is selected as the target routing scheme.

[0009] In some embodiments, when the preset service strategy is a balancing strategy, calculating a comprehensive score for each candidate routing scheme according to preset weight parameters, and selecting the candidate routing scheme with the highest comprehensive score as the target routing scheme includes: The payment success probability of the candidate routing scheme is multiplied by the first weight parameter to obtain the first weighted value; Based on the risk control trigger probability of the candidate routing scheme, the risk avoidance probability is calculated, and the risk avoidance probability is multiplied by the second weight parameter to obtain the second weighted value; The fee rate of the candidate routing scheme is multiplied by the third weight parameter to obtain the third weighted value; The first weighted value is added to the second weighted value, and then the third weighted value is subtracted to obtain the comprehensive score.

[0010] In some embodiments, processing the target payment request based on the target routing scheme includes: The target routing scheme is validated for compliance according to preset business rules; When the compliance verification passes, the target payment request is processed based on the target routing scheme; When processing the target payment request based on the target routing scheme fails, a candidate routing scheme that is different from the target routing scheme is selected from the plurality of candidate routing schemes for retry.

[0011] To achieve the above objectives, a second aspect of this application proposes a payment routing device based on deep learning, the device comprising: The acquisition module is used to acquire the target payment request and the channel characteristics of multiple alternative payment channels; The prediction module is used to predict the payment success probability and risk control trigger probability of each of the candidate payment channels based on the transaction characteristics of the target payment request and the channel characteristics using a pre-trained payment prediction model. The first determining module is used to determine multiple candidate routing schemes by taking the payment success probability, the risk control trigger probability, and the preset handling fee rate of each of the alternative payment channels as optimization targets and using a preset multi-objective optimization algorithm. The second determining module determines the target routing scheme from the multiple candidate routing schemes according to a preset business strategy. The processing module is used to process the target payment request based on the target routing scheme.

[0012] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0013] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0014] This application proposes a deep learning-based payment routing method and related equipment. The method first obtains the target payment request and the channel characteristics of multiple candidate payment channels; then, using a pre-trained payment prediction model, it predicts the payment success probability and risk control trigger probability of each candidate payment channel based on the transaction characteristics and channel characteristics of the target payment request; finally, it uses the payment success probability, risk control trigger probability, and the preset handling fee rate of each candidate payment channel as optimization objectives, and determines multiple candidate routing schemes through a preset multi-objective optimization algorithm; then, according to a preset business strategy, it determines the target routing scheme from the multiple candidate routing schemes; and finally, it processes the target payment request based on the target routing scheme.

[0015] According to the deep learning-based payment routing method and related equipment proposed in this application, the transaction characteristics of the target payment request and the channel characteristics of each payment channel are first obtained, providing a real-time data foundation for subsequent dynamic and personalized predictions. Next, a pre-trained payment prediction model is used to quantitatively predict the probability of payment success and the probability of risk control triggering, enabling routing decisions to shift from relying on static rules to being based on dynamic data predictions. Subsequently, the prediction results and transaction fees are used as optimization objectives, and a multi-objective optimization algorithm generates multiple candidate routing schemes. This step comprehensively weighs multiple business objectives, providing an optimal solution set for routing selection. Then, based on a preset business strategy, the final target routing scheme is determined from the candidate schemes and executed, ensuring that the final decision maintains optimality while meeting business needs in different scenarios. This application can dynamically generate personalized routing decisions for payment requests based on the real-time status of payment channels and the characteristics of users and merchants, thereby improving the success rate of payment transactions and the efficiency of routing selection.

[0016] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objectives and other advantages of this disclosure may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0017] Figure 1 This is a flowchart of the deep learning-based payment routing method provided in an embodiment of this application; Figure 2 This is another flowchart of the deep learning-based payment routing method provided in the embodiments of this application; Figure 3 This is another flowchart of the deep learning-based payment routing method provided in the embodiments of this application; Figure 4 This is another flowchart of the deep learning-based payment routing method provided in the embodiments of this application; Figure 5 This is another flowchart of the deep learning-based payment routing method provided in the embodiments of this application; Figure 6 This is another flowchart of the deep learning-based payment routing method provided in the embodiments of this application; Figure 7 This is a schematic diagram of the structure of the payment routing device based on deep learning provided in an embodiment of this application; Figure 8 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0021] In e-commerce and online services, the payment routing system is the core hub connecting merchants, users, and multiple third-party payment channels. Its main task is to select the optimal execution path for each payment request to ensure payment success rate, reduce transaction costs, and control risks.

[0022] In related technologies, traditional payment routing systems mostly adopt strategies based on static rules. For example, fixed channel priorities are preset for different businesses, and when the preferred channel fails, backup channels are tried in a predetermined order. Although such systems are simple to implement, as the scale of business expands and payment scenarios become more complex, their rigid routing strategies and lack of dynamic awareness of the real-time status of channels make it difficult to achieve personalized routing for different users and merchants, resulting in a low overall transaction success rate.

[0023] Based on this, embodiments of this application provide a payment routing method and related equipment based on deep learning, which can dynamically generate personalized routing decisions for payment requests based on the real-time status of the payment channel and the characteristics of users and merchants, thereby improving the success rate of payment transactions and the efficiency of routing selection.

[0024] The payment routing method and related devices based on deep learning provided in this application are specifically described through the following embodiments. First, the payment routing method based on deep learning in this application embodiment is described.

[0025] The deep learning-based payment routing method provided in this application is applicable to the fields of fintech, healthcare, and artificial intelligence. This method can be applied to terminals, servers, or software running on either. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the deep learning-based payment routing method, but is not limited to the above forms.

[0026] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0027] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0028] Figure 1 This is an optional flowchart of the deep learning-based payment routing method provided in the embodiments of this application. Figure 1The method may include, but is not limited to, steps S101 to S105.

[0029] Step S101: Obtain the target payment request and the channel characteristics of multiple alternative payment channels.

[0030] Step S102: Using a pre-trained payment prediction model, based on the transaction characteristics and channel characteristics of the target payment request, predict the payment success probability and risk control trigger probability of each alternative payment channel.

[0031] Step S103: Using the payment success probability, risk control trigger probability, and the preset handling fee rate of each alternative payment channel as optimization objectives, multiple candidate routing schemes are determined through a preset multi-objective optimization algorithm.

[0032] Step S104: Determine the target routing scheme from multiple candidate routing schemes according to the preset service policy.

[0033] Step S105: Process the target payment request based on the target routing scheme.

[0034] Steps S101 to S105, as illustrated in this embodiment, firstly, by acquiring the transaction characteristics of the target payment request and the channel characteristics of each payment channel, a real-time data foundation is provided for subsequent dynamic and personalized predictions. Next, a pre-trained payment prediction model is used to quantitatively predict the probability of payment success and the probability of risk control triggering, enabling routing decisions to shift from relying on static rules to being based on dynamic data predictions. Subsequently, the prediction results and transaction fees are used together as optimization objectives, and multiple candidate routing schemes are generated through a multi-objective optimization algorithm. This step comprehensively weighs multiple business objectives, providing a scientifically optimal solution set for routing selection. Then, based on a preset business strategy, the final target routing scheme is determined from the candidate schemes and executed, ensuring that the final decision maintains optimality while meeting business needs in different scenarios. This application can dynamically generate personalized routing decisions for payment requests based on the real-time status of payment channels and the characteristics of users and merchants, thereby improving the success rate of payment transactions and the efficiency of routing selection.

[0035] In step S101 of some embodiments, when a user initiates a payment, a target payment request is generated and delivered. This target payment request includes not only basic transaction information such as the transaction amount and timestamp, but also data such as the user's device information, IP address, and the merchant's industry category. Simultaneously, the application retrieves channel characteristics from various alternative payment channels, such as bank A, payment platform B, and e-wallet C, in real time. Channel characteristics are a set of data describing the current performance and historical attributes of a channel, specifically divided into two dimensions: first, historical channel characteristics, such as the average success rate of the channel in processing similar merchant transactions over the past month; and second, real-time channel characteristics, which are crucial for capturing the channel's instantaneous health, such as network latency and the number of transaction failures within the past five minutes. This step, by integrating the static information of the request with the dynamic information of the channel, constructs a comprehensive, multi-dimensional data snapshot for subsequent intelligent decision-making.

[0036] Please see Figure 2 In some embodiments, step S101 may include, but is not limited to, steps S101 to S104.

[0037] Step S201: Extract transaction features, user features, and merchant features from the target payment request.

[0038] Step S202: Obtain the historical channel characteristics and real-time channel characteristics of each alternative payment channel.

[0039] Step S203: Identify the user's geographic region in the user characteristics and the merchant's industry in the merchant characteristics as category features.

[0040] Step S204: Normalize the transaction amount in the transaction features, the historical success rate in the historical channel features, and the network latency in the real-time channel features, and use them as numerical features.

[0041] In step S201 of some embodiments, the received target payment request undergoes multi-source information aggregation to extract transaction features, user features, and merchant features that constitute its core information. Transaction features are stateless information directly parsed from the payment request data packet, such as a transaction amount of 500 yuan. User features and merchant features are state information that needs to be queried in real time. To ensure low latency, they are usually obtained by querying a pre-built feature storage implemented using an in-memory database such as Redis, using the user ID and merchant ID in the request. For example, the retrieved user features may include their membership level (VIP) and the number of recent payment failures (3); the retrieved merchant features may include their industry category (catering) and their historical average transaction amount (2500 yuan). The technical essence of this step is to enrich an instantaneous business behavior into an initial feature vector containing multi-dimensional information such as the transaction subject and transaction behavior through data aggregation.

[0042] In step S202 of some embodiments, attribute data for all candidate payment channels can be acquired in parallel. This data includes historical channel characteristics and real-time channel characteristics, which are combined to provide a channel evaluation that combines stability and sensitivity. Historical channel characteristics are lagging indicators reflecting the long-term performance of a channel, typically derived from a data warehouse updated daily by batch processing jobs. For example, it can be retrieved that channel A has an average success rate of 99.6% over the past 30 days. Real-time channel characteristics are forward-looking indicators reflecting the current instantaneous health of a channel, typically obtained by querying a monitoring dashboard calculated in real time by a stream processing engine. For example, it can be retrieved that channel A has an API call error rate of 0.01% and a current transaction queue depth of 15 over the past minute. Historical characteristics provide a baseline for decision-making, while real-time characteristics allow for instantaneous responses to sudden channel failures or congestion.

[0043] In step S203 of some embodiments, the feature set integrated in the previous steps is type-identified, and category features are separated according to a predefined feature outline. Category features refer to features whose values ​​are discrete, non-numerical labels, such as user region or merchant industry. The technical purpose of this step is to identify these text labels that cannot be directly involved in mathematical operations, preparing for subsequent encoding conversion. This preparation is crucial because the subsequent encoding process (e.g., one-hot encoding) will convert these labels into algebraic vectors. This conversion can prevent the prediction model from incorrectly interpreting ordinal relationships that do not exist between categories (e.g., incorrectly assuming that the catering industry is numerically greater than the retail industry). At the same time, the vocabulary management of features is also reflected in this step. Any merchant not in the predefined industry list can be classified into the "other" category. For example, when processing a payment request in real time, its merchant industry feature value, such as retail, will be queried in this vocabulary. If a match is found, its corresponding integer index, i.e., 1, is returned. If a new industry not seen during training is encountered, such as metaverse, the query will fail, and the process will map it to the index of the reserved [UNK] category. This mechanism ensures the robustness of the processing flow, avoids program crashes due to unknown inputs, and enables the model to learn a general representation for all unknown categories.

[0044] In step S204 of some embodiments, all numerical data needs to be normalized. Numerical features, such as transaction amount 500, historical success rate 0.996, and network latency 52, although numerical in themselves, have vastly different scales or dimensions. This difference can lead to gradient updates being disproportionately dominated by larger-scale features when training models using algorithms such as gradient descent, potentially causing model convergence difficulties or biases. This step maps all values ​​to a uniform [0,1] interval using a fixed scaling function, such as minimum-maximum scaling. It is worth noting that the minimum and maximum scaling parameters are statistically derived and saved from the entire training dataset during the offline training phase of the model. During online prediction, these fixed parameters are used to process real-time data, thereby ensuring consistency between online and offline processing logic.

[0045] Through steps S201 to S204, this embodiment first extracts the user, merchant, and transaction features of the request itself, and combines them with the historical and real-time features of the channel to construct a heterogeneous feature set that integrates transaction subjects, transaction behaviors, and transaction media, possessing both long-term statistical and instantaneous volatility. Subsequently, by identifying and separating categorical features, it lays the foundation for the subsequent encoding process of converting symbolic data into algebraic vectors, ensuring that the model can correctly process non-metric information. Finally, by normalizing the numerical features, it solves the problem of inconsistent dimensions between different features, thereby avoiding convergence difficulties or model bias caused by gradient scale differences during parameter optimization. The entire process transforms a raw business request into a structured, numerical, and mathematically normalized feature matrix. This matrix not only provides a format-compatible input for the subsequent prediction model but also ensures the stability of model training and the reliability of prediction results from a data perspective.

[0046] In step S102 of some embodiments, the feature data obtained in step S101 is input into a pre-trained payment prediction model. The payment prediction model is a deep neural network trained on massive amounts of successful and failed transaction data. Transaction features, i.e., the structured information parsed from the target payment request, are combined with the features of each channel to form the model's input vector. The payment prediction model outputs two key quantitative indicators: one is the payment success probability, which characterizes the likelihood that the transaction can be successfully completed through a specific channel; the other is the risk control trigger probability, used to measure the likelihood that the transaction will be intercepted by the channel's risk control strategy. For example, for a large virtual goods transaction occurring overnight, the success probability of channel A might be predicted to be 99%, but the risk control trigger probability is also as high as 30%, while the success probability of channel B is slightly lower at 97%, but the risk control trigger probability is only 1%.

[0047] Please see Figure 3 In some embodiments, the payment prediction model includes a payment success probability prediction sub-model and a risk control trigger probability prediction sub-model. Step S102 may include, but is not limited to, steps S301 to S303.

[0048] Step S301: Combine the categorical features and numerical features into model input features.

[0049] Step S302: Input the model input features into the payment success probability prediction sub-model to generate the payment success probability.

[0050] Step S303: Input the model input features into the risk control trigger probability prediction sub-model to generate the risk control trigger probability.

[0051] In step S301 of some embodiments, the categorical features and numerical features processed in the previous steps are concatenated to form the final model input features. After recognition and vocabulary mapping, the categorical features have been converted into a set of integer indices, which are further converted into high-dimensional binary vectors using encoding techniques. This step concatenates this high-dimensional binary vector with the normalized numerical feature vector, forming a fixed-dimensional and numerically encoded vector.

[0052] In step S302 of some embodiments, the model input features generated in the previous step are input into the payment success probability prediction sub-model to generate the payment success probability. The payment success probability prediction sub-model is a pre-trained deep learning network containing a large number of parameters solidified through learning from historical data. The pre-training process of this sub-model first acquires massive amounts of historical payment transaction data containing actual payment success or failure results. Then, for each historical transaction data, a feature extraction and processing method completely consistent with the online process is applied to construct the first training input, while the actual payment success result (e.g., 1 for success, 0 for failure) is constructed as the first training label. In the training loop, based on the difference between the first training input and the first training label, a first predicted loss value is calculated using a loss function (e.g., cross-entropy loss), and based on this loss value, the parameters inside the sub-model are iteratively adjusted by an optimizer algorithm until the first predicted loss value is lower than a preset first loss value threshold.

[0053] In step S303 of some embodiments, the model input features are fed in parallel into another independent risk control trigger probability prediction sub-model dedicated to risk assessment. This sub-model is structurally similar to the payment success probability prediction sub-model, but its internal parameters are trained based on different business objectives. It learns patterns related to risky behavior in historical transactions. Therefore, when the model input features flow through this network, it outputs a floating-point number between 0 and 1 to quantify the probability that the current transaction violates risk control rules, i.e., the risk control trigger probability. Its pre-training process is similar to that of the success rate sub-model, but uses different data labels: first, risk control interception logs containing actual risk control trigger results are obtained; then, a second training input is constructed based on historical payment transaction data, and the actual risk control trigger results obtained from the risk control logs are used as the second training label; finally, based on the second training input and the second training label, a second prediction loss value is calculated and the model parameters are iteratively adjusted until the second prediction loss value is lower than a preset second loss value threshold.

[0054] Through steps S301 to S303 described above, this embodiment first combines preprocessed features of different types into a unified model input feature. Then, by feeding this input feature in parallel into two predictive sub-models with clearly defined objectives, each pre-trained with different result data, a static feature is mapped to two independent probability indicators: the probability of successful payment and the probability of risk control triggering. This process provides a set of independent input data for subsequent multi-objective optimization decision-making steps, enabling decisions to be based not on the original features, but on probabilistic predictions of future outcomes.

[0055] In step S103 of some embodiments, this step defines the payment success probability and risk control trigger probability output from the previous step, along with the pre-configured transaction fee rate for each channel, as three optimization objectives. These three objectives are then processed by a pre-defined multi-objective optimization algorithm. This algorithm does not simply perform a weighted sum of the three indicators, but rather explores the Pareto front to find a set of optimal equilibrium solutions. The final output of multiple candidate routing schemes is such a set of solutions. For example, the algorithm might generate three candidate routing schemes: Scheme 1 (Channel A) has the highest success rate; Scheme 2 (Channel B) has the lowest transaction fee rate; and Scheme 3 (Channel C) achieves the best balance between success rate, risk, and cost. This step transforms the complex, multi-dimensional decision-making problem into a clear list of candidate schemes containing multiple high-quality options.

[0056] Please see Figure 4 In some embodiments, step S103 may include, but is not limited to, steps S401 to S402.

[0057] Step S401 sets maximizing the probability of successful payment, minimizing the probability of risk control triggering, and minimizing the transaction fee rate as multiple optimization objectives.

[0058] Step S402: By comparing multiple alternative payment channels pairwise, multiple candidate routing schemes are screened and determined. Each candidate routing scheme satisfies the following condition: no other alternative payment channel performs better in at least one optimization objective and performs no worse than the candidate routing scheme in all other optimization objectives.

[0059] In step S401 of some embodiments, the predicted payment success probability and risk control trigger probability from the preceding steps, along with the preset transaction fee rate for each channel, can be jointly established as multiple optimization objectives. This step transforms the problem of selecting the best channel into a multi-dimensional mathematical optimization problem. Specifically, the optimization objectives are defined as: maximizing the payment success probability to improve the transaction completion rate; minimizing the risk control trigger probability to avoid potential fraud or compliance risks; and minimizing the transaction fee rate to reduce the merchant's operating costs. These three objectives are inherently conflicting; for example, the channel with the highest success rate often has the highest transaction fee. Establishing these parallel, quantifiable optimization objectives provides a mathematical basis for subsequently finding the optimal balance point through algorithms.

[0060] In step S402 of some embodiments, the process uses an algorithm to filter and determine multiple candidate routing schemes, the core of which is to find a set of Pareto optimal solutions. The algorithm compares all candidate payment channels pairwise to determine their relative merits. Each candidate routing scheme must satisfy a strict condition: there is no other candidate payment channel that performs better on at least one optimization objective and is no worse on all other optimization objectives. For example, suppose there are three channels: channel A (99% success rate, 0.6% cost), channel B (98% success rate, 0.4% cost), and channel C (99% success rate, 0.5% cost). In this comparison, channel C is better than channel A because they have the same success rate, but C has a lower cost. However, there is no absolute superiority or inferiority between channel B and channel C, because C has a higher success rate, but B has a lower cost. Therefore, the final selected candidate routing schemes will be the set of channels B and C, while channel A is excluded.

[0061] Through steps S401 to S402 described above, this embodiment first transforms the routing decision problem into a multi-objective mathematical optimization problem by establishing success rate, risk, and cost as three parallel optimization objectives. Subsequently, it replaces the traditional method's single-score calculation based on subjective weighting by finding a set of Pareto optimal solutions where no absolutely better solution exists. The entire process decomposes a complex decision-making process into two steps: objective definition and optimal solution set finding. The output set of candidate routing schemes provides mathematical verification for subsequent selection based on specific business strategies, thereby improving the rationality of the decision.

[0062] In step S104 of some embodiments, a final selection is made from multiple candidate routing schemes generated in the previous step according to a preset business strategy. The business strategy is a decision rule pre-configured based on different business scenarios and operational goals. For example, during a large-scale promotional event on an e-commerce platform, to maximize user experience and transaction completion, the business strategy might be set to a success rate priority strategy. In this case, even if Scheme 1 (Channel A) has the highest transaction fee rate, it will be selected as the target routing scheme because of its highest payment success probability. Conversely, during normal operating hours, the business strategy might switch to a cost priority strategy, in which case Scheme 2 (Channel B) with the lowest transaction fee rate will be selected. This step ensures that the technically optimal solution remains consistent with changing business objectives.

[0063] Please see Figure 5 In some embodiments, step S104 may include, but is not limited to, steps S501 to S503.

[0064] Step S501: When the preset business strategy is a cost-first strategy, select the candidate routing scheme with the lowest transaction fee rate from multiple candidate routing schemes as the target routing scheme.

[0065] Step S502: When the preset business strategy is the success rate priority strategy, select the candidate routing scheme with the highest payment success probability from multiple candidate routing schemes as the target routing scheme.

[0066] Step S503: When the preset service strategy is a balanced strategy, a comprehensive score is calculated for each candidate routing scheme according to the preset weight parameters, and the candidate routing scheme with the highest comprehensive score is selected as the target routing scheme.

[0067] In step S501 of some embodiments, the core objective of the cost-first strategy is to minimize the direct transaction fee cost per transaction, which is typically applicable to business scenarios with low profit margins or strict cost control. The process iterates through all candidate routing schemes generated in previous steps and compares the transaction fee rate of each scheme. Ultimately, the candidate routing scheme with the lowest transaction fee rate is determined as the target routing scheme. For example, if the candidate scheme set includes channel B (98% success rate, 0.4% fee) and channel C (99% success rate, 0.5% fee), under this strategy, the application will choose channel B because its 0.4% fee rate is lower than channel C's 0.5%, even though its success rate is relatively lower.

[0068] In step S502 of some embodiments, the goal of the success rate priority strategy is to maximize the success rate of transactions, which is crucial for improving user experience and ensuring the smooth completion of critical transactions. The process will again traverse all candidate routing schemes, but this time the comparison is based on the payment success probability of each scheme. The candidate routing scheme with the highest predicted payment success probability will be selected as the target routing scheme. Taking the aforementioned candidate scheme set as an example, namely channel B (98% success rate, 0.4% fee) and channel C (99% success rate, 0.5% fee), under this strategy, the application will choose channel C because its 99% predicted success rate is higher than channel B's 98%. In this case, the relatively higher fee is considered a necessary cost to ensure the success rate.

[0069] In step S503 of some embodiments, when the preset business strategy is a balanced strategy, a refined decision-making process based on a utility function can be executed to seek an optimal solution after weighted trade-offs among multiple objectives. For each candidate routing scheme, a comprehensive scoring function is used to calculate its quantified business value. This comprehensive score is calculated by a preset linear combination function, and its mathematical expression can be defined as:

[0070] In this function, It represents the payment success probability of the candidate routing scheme. It is the probability of its risk control being triggered. It is its handling fee rate. , , These are preset weight parameters, which are optimized by business analysts based on long-term operational goals through simulation backtesting using historical data. They quantify the relative importance the business places on success rate, risk avoidance, and cost savings. The calculation process will be illustrated through a specific example. Assume the candidate routing scheme set contains two schemes: Channel B: Successful payment probability p=0.98, risk control trigger probability r=0.002, transaction fee rate f=0.004.

[0071] Channel C: Payment success probability p=0.99, risk control trigger probability r=0.001, transaction fee rate f=0.005.

[0072] Assume the currently effective weight parameters are set to a set of values ​​that highly value success rate but are also sensitive to cost: α=10, β=2, γ=80. The application will calculate a comprehensive score for each of the two channels: Channel B has a comprehensive score of 11.476, and Channel C has a comprehensive score of 11.498. Therefore, under this balanced strategy, the application will select Channel C, which has the highest comprehensive score, as the target routing solution. This step transforms multi-dimensional and difficult-to-compare candidate solutions into a quantifiable utility function relevant to business objectives, ultimately leading to a final decision.

[0073] Through steps S501 to S503 described above, this embodiment of the application first ensures that routing selection can achieve a single business objective in extreme scenarios by providing two clearly defined strategies: cost priority and success rate priority. Then, by providing a balanced strategy based on configurable weight parameters to calculate a comprehensive score, it enables trade-offs among multiple objectives. The entire process transforms the candidate solution set generated in the preceding steps into a target routing solution consistent with the current business strategy, thereby achieving a shift from technical optimization to business optimization.

[0074] In step S105 of some embodiments, after the target routing scheme is determined, the specific payment channel included in it is used to process the target payment request. The relevant data of the target payment request, such as the order number, amount, and payee account, is encapsulated and sent to the application programming interface (API) of the target channel. Furthermore, this step may include fault tolerance and degradation mechanisms. If the first attempt based on the target routing scheme fails, for example due to a temporary channel failure or network timeout, the process will not immediately terminate and return a failure result to the user. Instead, it will automatically select the next best solution from the candidate routing scheme list generated in step S103 for a degradation retry. This design enhances the stability of the payment process and improves the final transaction success rate.

[0075] Please see Figure 6 In some embodiments, step S105 may include, but is not limited to, steps S601 to S603.

[0076] Step S601: Perform compliance verification on the target routing scheme according to preset business rules.

[0077] Step S602: When the compliance verification passes, process the target payment request based on the target routing scheme.

[0078] Step S603: When processing the target payment request based on the target routing scheme fails, select a candidate routing scheme that is different from the target routing scheme from multiple candidate routing schemes and retry.

[0079] In step S601 of some embodiments, a final compliance check is performed before the payment request is sent to the external channel. The purpose of this step is to enforce a set of business rules defined by the business onto the target routing scheme. These business rules are independent of the prediction model and typically exist in the form of a configurable rule set. For example, one rule might stipulate that channel A is prohibited for transactions exceeding 50,000 yuan; another rule might stipulate that merchants in the gaming industry are not allowed to receive payments through channel B. When a payment request of 60,000 yuan has channel A as its target routing scheme, the verification logic in this step will determine that the scheme is non-compliant and reject it, thereby triggering a reselection process.

[0080] In step S602 of some embodiments, the actual payment request processing is only executed after the target routing scheme passes the compliance verification in the previous step. Specifically, the process retrieves information such as the API access address, merchant ID, and encryption key of the channel specified in the target routing scheme from the configuration library. Then, the order number, amount, currency, and other data from the original payment request are assembled into a standard data request packet according to the interface documentation specifications of the specific channel, and sent to the channel's server via an encrypted HTTP connection. Afterward, the application enters a brief waiting state to receive a processing receipt from the channel.

[0081] In step S603 of some embodiments, when the processing based on the target routing scheme in step S602 fails, for example, if the channel returns a temporary error such as "system busy" or a network connection timeout occurs, unlike traditional routing which tries the next backup channel in a fixed order, the retry logic here returns to the set of multiple candidate routing schemes generated in the previous step. From this set, a candidate routing scheme that is different from the target routing scheme that just failed and is rated as the second best according to the current business strategy is selected. For example, if the preferred scheme C with the highest success rate fails due to timeout, this step will automatically select the second highest success rate scheme B among the candidate schemes and initiate a completely new payment request processing.

[0082] Through steps S601 to S603 described above, this embodiment first introduces compliance verification based on business rules to add security business constraints to the probabilistic decision-making of the model. After successful verification, the payment request is executed, transforming the routing decision into an actual transaction action. Finally, by employing a mechanism that dynamically retryes suboptimal solutions selected from the set of optimal candidate solutions, the final success rate of the payment process in the face of temporary channel failures is improved. The entire process ensures that the intelligent routing decision is both compliant and stable and reliable during execution.

[0083] Please see Figure 7This application also provides a deep learning-based payment routing device that can implement the above-described deep learning-based payment routing method. The device includes: The acquisition module is used to acquire the target payment request and the channel characteristics of multiple alternative payment channels; The prediction module is used to predict the payment success probability and risk control trigger probability of each of the candidate payment channels based on the transaction characteristics of the target payment request and the channel characteristics using a pre-trained payment prediction model. The first determining module is used to determine multiple candidate routing schemes by taking the payment success probability, the risk control trigger probability, and the preset handling fee rate of each of the alternative payment channels as optimization targets and using a preset multi-objective optimization algorithm. The second determining module determines the target routing scheme from the multiple candidate routing schemes according to a preset business strategy. The processing module is used to process the target payment request based on the target routing scheme.

[0084] The specific implementation of this deep learning-based payment routing device is basically the same as the specific implementation of the deep learning-based payment routing method described above, and will not be repeated here.

[0085] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned deep learning-based payment routing method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0086] Please see Figure 8 , Figure 8 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 801 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 802 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 802 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 802 and is called and executed by the processor 801 to implement the deep learning-based payment routing method of the embodiments of this application. The 803 input / output interface is used to implement information input and output. The communication interface 804 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 805 transmits information between various components of the device (e.g., processor 801, memory 802, input / output interface 803, and communication interface 804); The processor 801, memory 802, input / output interface 803, and communication interface 804 are connected to each other within the device via bus 805.

[0087] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned deep learning-based payment routing method.

[0088] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0089] This application proposes a deep learning-based payment routing method and related equipment. The method first obtains the target payment request and the channel characteristics of multiple candidate payment channels; then, using a pre-trained payment prediction model, it predicts the payment success probability and risk control trigger probability of each candidate payment channel based on the transaction characteristics and channel characteristics of the target payment request; finally, it uses the payment success probability, risk control trigger probability, and the preset handling fee rate of each candidate payment channel as optimization objectives, and determines multiple candidate routing schemes through a preset multi-objective optimization algorithm; then, according to a preset business strategy, it determines the target routing scheme from the multiple candidate routing schemes; and finally, it processes the target payment request based on the target routing scheme.

[0090] According to the deep learning-based payment routing method and related equipment proposed in this application, the transaction characteristics of the target payment request and the channel characteristics of each payment channel are first obtained, providing a real-time data foundation for subsequent dynamic and personalized predictions. Next, a pre-trained payment prediction model is used to quantitatively predict the probability of payment success and the probability of risk control triggering, enabling routing decisions to shift from relying on static rules to being based on dynamic data predictions. Subsequently, the prediction results and transaction fees are used as optimization objectives, and a multi-objective optimization algorithm generates multiple candidate routing schemes. This step comprehensively weighs multiple business objectives, providing an optimal solution set for routing selection. Then, based on a preset business strategy, the final target routing scheme is determined from the candidate schemes and executed, ensuring that the final decision maintains optimality while meeting business needs in different scenarios. This application can dynamically generate personalized routing decisions for payment requests based on the real-time status of payment channels and the characteristics of users and merchants, thereby improving the success rate of payment transactions and the efficiency of routing selection.

[0091] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0092] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0093] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0094] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0095] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0096] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0097] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0098] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0099] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0100] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A payment routing method based on deep learning, characterized in that, The method includes: Obtain the target payment request and the channel characteristics of multiple alternative payment channels; Using a pre-trained payment prediction model, based on the transaction characteristics of the target payment request and the channel characteristics, the success probability and risk control trigger probability of each of the alternative payment channels are predicted. The payment success probability, the risk control trigger probability, and the preset handling fee rate of each of the alternative payment channels are used as optimization objectives. Multiple candidate routing schemes are determined through a preset multi-objective optimization algorithm. Based on a preset business strategy, a target routing scheme is determined from the multiple candidate routing schemes; The target payment request is processed based on the target routing scheme.

2. The payment routing method based on deep learning according to claim 1, characterized in that, The characteristics of the target payment request and multiple alternative payment channels include: Extract transaction features, user features, and merchant features from the target payment request; Obtain the historical and real-time channel characteristics of each of the candidate payment channels; The user's geographic location in the user characteristics and the merchant's industry in the merchant characteristics are identified as category features; The transaction amount in the transaction features, the historical success rate in the historical channel features, and the network latency in the real-time channel features are normalized and used as numerical features.

3. The payment routing method based on deep learning according to claim 2, characterized in that, The payment prediction model includes a payment success probability prediction sub-model and a risk control trigger probability prediction sub-model. The pre-trained payment prediction model, based on the transaction characteristics of the target payment request and the channel characteristics, predicts the payment success probability and risk control trigger probability of each of the candidate payment channels, including: The categorical features and the numerical features are combined into model input features; The input features of the model are input into the payment success probability prediction sub-model to generate the payment success probability; wherein, the pre-training process of the payment success probability prediction sub-model includes: acquiring historical payment transaction data containing actual payment success results; constructing a first training input based on the historical payment transaction data, and constructing the actual payment success results as a first training label; and, based on the first training input and the second training label, calculating a first prediction loss value and iteratively adjusting the model parameters of the payment success probability prediction sub-model until the first prediction loss value is lower than a preset first loss value threshold. The model input features are input into the risk control trigger probability prediction sub-model to generate the risk control trigger probability; wherein, the pre-training process of the risk control trigger probability prediction sub-model includes: obtaining risk control interception logs containing actual risk control trigger results; constructing a second training input based on historical payment transaction data, and constructing the actual risk control trigger results as a second training label; and, based on the second training input and the second training label, calculating a second prediction loss value and iteratively adjusting the model parameters of the risk control trigger probability prediction sub-model until the second prediction loss value is lower than a preset second loss value threshold.

4. The payment routing method based on deep learning according to claim 1, characterized in that, The method uses the payment success probability, the risk control trigger probability, and the preset handling fee rate of each of the candidate payment channels as optimization objectives, and determines multiple candidate routing schemes through a preset multi-objective optimization algorithm, including: Maximizing the payment success probability, minimizing the risk control trigger probability, and minimizing the transaction fee rate are taken as multiple optimization objectives; By comparing the multiple alternative payment channels pairwise, multiple candidate routing schemes are screened and determined. Each candidate routing scheme satisfies the following condition: no other alternative payment channel performs better than the candidate routing scheme in at least one of the optimization objectives, and performs no worse than the candidate routing scheme in all other optimization objectives.

5. The payment routing method based on deep learning according to claim 1, characterized in that, The preset business strategies include: cost-first strategy, success-rate-first strategy, and balancing strategy. The step of determining the target routing scheme from the multiple candidate routing schemes according to the preset business strategies includes: When the preset business strategy is a cost-first strategy, the candidate routing scheme with the lowest transaction fee rate is selected from the multiple candidate routing schemes as the target routing scheme; When the preset business strategy is a success rate priority strategy, the candidate routing scheme with the highest payment success probability is selected from the multiple candidate routing schemes as the target routing scheme; When the preset service strategy is a balanced strategy, a comprehensive score is calculated for each candidate routing scheme according to the preset weight parameters, and the candidate routing scheme with the highest comprehensive score is selected as the target routing scheme.

6. The payment routing method based on deep learning according to claim 5, characterized in that, When the preset service strategy is a balanced strategy, a comprehensive score is calculated for each candidate routing scheme according to preset weight parameters, and the candidate routing scheme with the highest comprehensive score is selected as the target routing scheme, including: The payment success probability of the candidate routing scheme is multiplied by the first weight parameter to obtain the first weighted value; Based on the risk control trigger probability of the candidate routing scheme, the risk avoidance probability is calculated, and the risk avoidance probability is multiplied by the second weight parameter to obtain the second weighted value; The fee rate of the candidate routing scheme is multiplied by the third weight parameter to obtain the third weighted value; The first weighted value is added to the second weighted value, and then the third weighted value is subtracted to obtain the comprehensive score.

7. The payment routing method based on deep learning according to claim 1, characterized in that, The process of processing the target payment request based on the target routing scheme includes: The target routing scheme is validated for compliance according to preset business rules; When the compliance verification passes, the target payment request is processed based on the target routing scheme; When processing the target payment request based on the target routing scheme fails, a candidate routing scheme that is different from the target routing scheme is selected from the plurality of candidate routing schemes for retry.

8. A payment routing device based on deep learning, characterized in that, The device includes: The acquisition module is used to acquire the target payment request and the channel characteristics of multiple alternative payment channels; The prediction module is used to predict the payment success probability and risk control trigger probability of each of the candidate payment channels based on the transaction characteristics of the target payment request and the channel characteristics using a pre-trained payment prediction model. The first determining module is used to determine multiple candidate routing schemes by taking the payment success probability, the risk control trigger probability, and the preset handling fee rate of each of the alternative payment channels as optimization targets and using a preset multi-objective optimization algorithm. The second determining module determines the target routing scheme from the multiple candidate routing schemes according to a preset business strategy. The processing module is used to process the target payment request based on the target routing scheme.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the deep learning-based payment routing method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, characterized in that, when the computer program is executed by a processor, it implements the deep learning-based payment routing method according to any one of claims 1 to 7.

Images