Cross-border payment path optimization system and method based on multi-dimensional data analysis

Abstract

The invention relates to the technical field of big data processing and analysis, in particular to a cross-border payment path optimization system and method based on multi-dimensional data analysis. Payment transaction data and exchange rate data are collected, a multi-dimensional payment path evaluation model is constructed from the angles of payment cost, payment timeliness, payment risk, payment availability and the like, and each payment path is scored in real time; constructing a payment path optimization model, dynamically selecting an optimal path and performing intelligent scheduling; a machine learning algorithm is introduced, an abnormal transaction mode is automatically identified from mass payment data, and a real-time anomaly detection and risk prevention and control strategy is established; according to the method provided by the invention, the success rate of cross-border payment can be improved, the cost is reduced, the payment time is shortened, and the payment risk is prevented, so that the cross-border payment service quality and the user experience are comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing and analysis, and specifically to a cross-border payment path optimization system and method based on multidimensional data analysis. Background Art

[0002] In the existing technology, cross-border payment path optimization mainly adopts methods such as heuristic algorithms and linear programming. These methods require pre-setting of optimization goals and constraints, and rely on expert experience to manually adjust parameters, making it difficult to achieve real-time dynamic optimization. In addition, most existing methods construct evaluation functions based on a single dimension (such as cost or time), lacking comprehensive consideration of multiple key factors such as payment risk and availability. In view of the complex business scenarios and risk control and compliance requirements faced by cross-border payments, single-dimensional path evaluation often fails to meet actual needs.

[0003] The patent application with publication number CN113313479A discloses a payment business big data processing method and system based on artificial intelligence. Through description optimization and quantitative expression, the completion index of the data feature processing unit can be flexibly determined. On the one hand, for the pending cross-border payment data with a low description optimization range, only a small number of feature weighting modules are needed to obtain better feature processing results, which can avoid key content deviations caused by additional and unnecessary feature operations. On the other hand, for the pending cross-border payment data with excessive description optimization, further feature weighting processing can be used to ensure the content fidelity of the pending cross-border payment data and improve the traceability and recovery performance of the pending cross-border payment data. Therefore, through automated and intelligent solutions, the description optimization elimination effect and the processing errors caused by key content deviations are balanced to a certain extent, and the resource overhead of the payment business big data processing system can be saved, and the processing efficiency of cross-border payment data can be improved.

[0004] In terms of payment anomaly detection, existing methods face the ever-changing business environment and user behavior. Solidified rules are often difficult to effectively deal with new fraud methods, and the rule formulation and update cycle is long and the timeliness is poor. The rule engine method has a high rate of missed reports and false positives, which increases the cost of manual review on the one hand and affects the user experience on the other. Therefore, it is necessary to introduce machine learning technology to automatically mine abnormal patterns from massive historical transaction data, build an adaptive anomaly detection model, and realize real-time risk identification and blocking.

[0005] Existing methods lack flexibility and are difficult to adapt to increasingly complex cross-border payment scenarios. There is an urgent need to introduce big data analysis and machine learning technologies to fully tap the value of payment data and achieve intelligent optimization of payment paths and precise prevention and control of risks.

[0006] In view of this, the present invention proposes a cross-border payment path optimization system and method based on multi-dimensional data analysis. Summary of the invention

[0007] To achieve the above objectives, the present invention provides a cross-border payment path optimization method based on multidimensional data analysis, and the specific technical solution is as follows:

[0008] Collect payment path data, collect payment transaction data and exchange rate data by parsing payment gateway logs and calling APIs, and perform data cleaning;

[0009] Construct a multi-dimensional payment path evaluation model from the perspectives of payment cost, payment timeliness, payment risk, and payment availability, and perform real-time scoring on various payment path indicators;

[0010] Build a payment path optimization model, dynamically select the optimal path based on the real-time scoring of the multi-dimensional payment path evaluation model, and perform path scheduling based on business rules and concurrent request factors;

[0011] Establish payment path anomaly detection and risk control strategies, automatically identify abnormal transaction patterns from massive payment data through machine learning algorithms, and block high-risk transactions in real time.

[0012] Preferably, a payment log file is extracted from a payment gateway of a cross-border payment path, and for each log file, payment data in the log file is extracted and converted into structured transaction data; the data structure of a single transaction record is defined as a tuple;

[0013] Determine the set of currency pairs for which exchange rates need to be collected. For each currency pair, call the exchange rate API to obtain the exchange rate time series over a period of time; match the exchange rate data with the transaction data, and associate the transaction records with the exchange rate;

[0014] Clean the original transaction data set, remove duplicate transaction records, and filter by transaction status to eliminate failed, abnormal, and test abnormal transactions; convert the transaction amount into the amount of the base currency according to the exchange rate; sort by time to obtain the transaction time series; generate the cleaned transaction data set;

[0015] Based on the cleaned transaction data set, aggregate by channel, scenario and time dimension to generate summary statistics of different granularities; define statistical indicators, including total transaction amount, average arrival time and arrival success rate;

[0016] Calculates the specified payment path statistical indicator value.

[0017] Preferably, for each path in the payment path set, based on the statistical feature vector of the payment path, an evaluation index is constructed from multiple dimensions of payment cost, payment timeliness, payment risk, and payment availability:

[0018] The payment cost refers to the transaction cost of the path, including the handling fee rate and exchange loss rate of each channel and node; the payment time refers to the end-to-end processing time of the path, including forwarding delay, queuing waiting time and cross-border settlement time; the payment risk refers to the potential risk of the path, taking into account the risk of fund theft, counterparty default risk and system failure risk; the payment availability refers to the technical availability of the path, taking into account the impact of node failure and network interruption;

[0019] The evaluation indicators are constructed based on multiple dimensions to construct a comprehensive utility score of the path.

[0020] Preferably, based on the constructed payment path evaluation model, a real-time payment path optimization model is constructed using machine learning technology;

[0021] Construct training samples. Based on historical transaction data and real-time scoring of the evaluation model, construct training samples for supervised learning; randomly divide the training samples into training sets and test sets;

[0022] Design a multi-classification machine learning model with transaction features and path sets as input and matching scores of each path as output; consider the heterogeneity of transaction features and path features and use a multi-layer perceptron (MLP) model for fusion modeling;

[0023] Based on the training set samples, the model parameters are optimized using the cross entropy loss function and the stochastic gradient descent algorithm, and the model parameters are updated using the back propagation algorithm until the loss function converges or the preset number of iterations is reached;

[0024] Based on the test set samples, the classification accuracy, precision, and recall of the model are evaluated, hyperparameters are tuned, and the model with the best performance is selected for optimal path prediction.

[0025] Preferably, for transaction requests arriving in real time, their feature vectors are extracted, a set of candidate paths is generated, and the sets are substituted into the trained model to calculate the matching scores;

[0026] Select the path with the highest matching degree as the optimal path, and forward the request to the channel corresponding to the path for processing;

[0027] A path scheduling mechanism is introduced to record the historical load level for each channel. When the load level exceeds the preset warning line, the transaction requests arriving at the channel are smoothed and valley-filled, and are instead allocated to the suboptimal path for processing until the channel load returns to normal levels.

[0028] Preferably, constructing abnormal transaction feature engineering to extract feature vectors reflecting transaction abnormality from payment gateway logs and transaction data sets;

[0029] Based on the transaction record data, a multi-dimensional abnormal feature vector is constructed; an abnormal transaction detection model is constructed, and based on the abnormal transaction features, a supervised learning classification model is constructed to detect whether the transaction data has abnormal transaction features.

[0030] Preferably, the trained anomaly detection model is deployed to the payment gateway, anomaly feature vectors are extracted from real-time transaction requests, and input into the model for real-time prediction to obtain anomaly probability;

[0031] Set an abnormality judgment threshold. When the abnormal probability is greater than the abnormality judgment threshold, the transaction is judged as abnormal, and a rejection strategy is implemented to block the transaction request.

[0032] Design a graded blocking strategy, set multi-level risk thresholds, divide the risk into three levels: low, medium and high according to the abnormal probability prediction value, and take corresponding disposal measures according to the risk level.

[0033] A cross-border payment path optimization system based on multi-dimensional data analysis, which is used to implement the cross-border payment path optimization method based on multi-dimensional data analysis, including: a data acquisition module, a multi-dimensional payment path evaluation module, a payment path optimization module and a payment risk control module;

[0034] The data collection module is used to collect payment path data, collect payment transaction data and exchange rate data by parsing payment gateway logs and calling APIs, and perform data cleaning;

[0035] The multi-dimensional payment path evaluation module is used to construct a multi-dimensional payment path evaluation model, build an evaluation model from the perspectives of payment cost, payment timeliness, payment risk and payment availability, and perform real-time scoring on various indicators of the payment path;

[0036] The payment path optimization module is used to build a payment path optimization model, dynamically select the optimal path based on the real-time scoring of the multi-dimensional payment path evaluation model, and perform path scheduling in combination with business rules and concurrent request factors;

[0037] The payment risk control module is used to establish payment path anomaly detection and risk control strategies, automatically identify abnormal transaction patterns from massive payment data through machine learning algorithms, and block high-risk transactions in real time.

[0038] An electronic device comprises: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; the processor executes the cross-border payment path optimization method based on multidimensional data analysis by calling the computer program stored in the memory.

[0039] A computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the cross-border payment path optimization method based on multidimensional data analysis.

[0040] The beneficial effects of the present invention are as follows: the present invention collects payment path data and performs data cleaning to provide a high-quality data basis for subsequent model construction, reduce the interference of noise and abnormal data, and improve the accuracy and robustness of the model.

[0041] The present invention constructs a multi-dimensional payment path evaluation model, comprehensively considers multiple key factors such as payment cost, timeliness, risk and availability, realizes a comprehensive evaluation of the payment path, and provides a scientific basis for path optimization.

[0042] The present invention constructs a payment path optimization model, dynamically selects the optimal path based on real-time scoring, and performs scheduling in combination with business rules and concurrent factors, thereby improving payment success rate, reducing payment costs, and improving user experience.

[0043] The present invention establishes payment path anomaly detection and risk control strategies, utilizes machine learning to automatically identify abnormal transaction patterns, realizes real-time risk prevention and control, reduces manual review costs, improves risk control timeliness, and ensures payment security. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of the cross-border payment path optimization method based on multidimensional data analysis provided by the present invention;

[0045] Figure 2 This is a structural diagram of the cross-border payment path optimization system based on multidimensional data analysis provided by the present invention. DETAILED DESCRIPTION

[0046] In order to better understand the present invention, a more detailed description will be made of various aspects of the present invention with reference to the accompanying drawings. It should be understood that these detailed descriptions are only descriptions of exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0047] In the accompanying drawings, the size, dimensions and shapes of the elements have been slightly adjusted for ease of illustration. The accompanying drawings are for illustration only and are not strictly drawn to scale. As used herein, the terms "substantially", "approximately" and similar terms are used as terms of approximation, not as terms of degree, and are intended to illustrate the inherent deviations in measurements or calculations that will be recognized by those of ordinary skill in the art. In addition, in the present invention, the order in which the steps are described does not necessarily represent the order in which these processes occur in actual operation, unless otherwise specified or can be derived from the context.

[0048] It should also be understood that expressions such as "comprises", "including", "having", "includes" and / or "comprising" are open rather than closed expressions in this specification, which indicate the presence of the stated features, elements and / or components, but do not exclude the presence of one or more other features, elements, components and / or combinations thereof. In addition, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features rather than just the individual elements in the list. In addition, when describing embodiments of the present invention, "may" is used to mean "one or more embodiments of the present invention". And, the term "exemplary" is intended to refer to an example or illustration.

[0049] Unless otherwise defined, all words (including engineering terms and scientific and technological terms) used in this article have the same meaning as those commonly understood by ordinary technicians in the field to which the present invention belongs. It should also be understood that unless there is a clear explanation in the present invention, the words defined in the commonly used dictionary should be interpreted as having the same meaning as their meaning in the context of the relevant technology, and should not be interpreted in an idealized or overly formal sense.

[0050] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0051] Example 1

[0052] Reference Figure 1 , which is the first embodiment of the present invention, provides a cross-border payment path optimization method based on multidimensional data analysis.

[0053] S1: Collect payment path data, collect payment transaction data and exchange rate data by parsing payment gateway logs and calling APIs, and perform data cleaning.

[0054] Extract payment log files from the payment gateway of the cross-border payment path. For each log file, extract the payment data in the log file and convert it into structured transaction data; define the data structure of a single transaction record as a tuple:

[0055] y=(id,t,src,dst,A,cur,status);

[0056] Among them, the tuple y represents the transaction id at time t, sending the amount of A in the currency cur from the node of the src channel to the node of the dst channel, and the transaction status is status; all the transaction records obtained by the analysis form a set Y = y 1 ,y 2 ,...,y B , B is the total number of transactions.

[0057] Determine the currency pair set Q for which exchange rates need to be collected = (cur 1 ,cur 2 )|cur 1 ,cur 2 ∈CUR, where CUR is the set of all currencies; for each currency pair (cur 1 ,cur 2 )∈Q, call the exchange rate API to obtain the exchange rate time series over a period of time Exchange rate time series represents time t 1 Exchange Cur 2 exchange rate; matching exchange rate data with transaction data, and recording transaction records Linked exchange rate where cur base The base currency.

[0058] Clean the original transaction data set Y, remove duplicate transaction records, and filter by transaction status to eliminate failed, abnormal, and test abnormal transactions; convert the transaction amount A into the base currency cur at the exchange rate r base The amount A'=A·r, eliminating currency differences; sorting by time to obtain the transaction time series; generating the cleaned transaction data set Y'.

[0059] Based on the cleaned transaction data set Y', aggregate by channel, scenario and time dimensions to generate summary statistics of different granularities; define the statistical indicator stat j , including total transaction amount, average arrival time and arrival success rate.

[0060] For a specified payment path p∈P, the transaction record Y p =y∈Y'|y.src∈p∧y.dst∈p to extract and calculate the statistical index value v pj =f j (y p ), where f jis the indicator calculation function; get the statistical characteristic vector v of the payment path p p =(v p1 ,v p2 ,...,v pJ ), j∈[1,J], as the input for evaluation optimization.

[0061] Step S1 establishes a complete, accurate, and fine-grained multi-dimensional transaction data set by collecting and cleaning the original transaction data and exchange rate data of the payment path, providing a high-quality data foundation for subsequent payment path evaluation, optimization, anomaly detection and other tasks.

[0062] S2: Construct a multi-dimensional payment path evaluation model from the perspectives of payment cost, payment timeliness, payment risk, and payment availability, and perform real-time scoring on various payment path indicators.

[0063] For each path p in the payment path set P, based on the statistical feature vector V of the payment path p p =(vp 1 , vp 2 , ..., vp J ), constructing evaluation indicators from multiple dimensions including payment cost, payment timeliness, payment risk and payment availability.

[0064] The payment cost indicator c p , represents the transaction cost of path p, including the handling fee rate f of each channel and node pj and exchange loss rate l pj , let the transaction amount of path p be A p , then the total cost is: c p =∑ j (f pj +l pj )·A p ; where f pj , l pj From the statistical characteristics v pj Obtained in.

[0065] The payment time index t p , represents the end-to-end processing time of path p, including the forwarding delay d pj , waiting time in queue pj and cross-border settlement times pj , then the total time is: t p =∑ j (d pj +q pj +s pj );where d pj ,q pj ,s pj From v pjextract.

[0066] The payment risk indicator r p , represents the potential risk of path p, considering the risk of funds being stolen e 1 、Counterparty default risk 2 and system failure risk 3 , let the probability of occurrence of the ith risk event be P(e i ), the average loss amount of a single event is L(e i ), the number of daily transactions of path p is N p , then the expected total risk is: Among them, the probability P(e i ) is given according to historical data and expert experience, the loss L(e i ) is determined by business rules, the number of transactions N p From v pj Obtained from statistics.

[0067] The payment availability indicator a p , represents the technical availability of path p, considering the impact of node failure and network interruption; let the failure rate of a single node be P n , the network interruption rate is P l , path p contains m nodes, the probability of path unavailability is:

[0068] P(E u )=1-(1-P n ) m (1-P l );

[0069] The availability index is the complementary probability a p =1-P(E u ), where the failure rate P n and the interruption rate P l Calculated using historical failure data.

[0070] Define the utility function u for the jth evaluation indicator j (·), considering the importance of different indicators, define the indicator weight vector w = (w 1 , w 2 , w 3 , w 4 ), then the comprehensive utility score of path p is:

[0071] U p =w 1 u 1 (c p )+w 2 u 2 (t p )+w3 u 3 (r p )+w 4 u 4 (a p );

[0072] Among them, the weights satisfy

[0073] Step S2 builds a comprehensive payment path evaluation model based on multiple dimensions such as payment cost, payment timeliness, payment risk and payment availability. The payment path evaluation model can quantify the performance of different paths in various key indicators, provide a basis for dynamically selecting the optimal path, and help improve the efficiency of cross-border payments and user experience.

[0074] S3: Build a payment path optimization model, dynamically select the optimal path based on the real-time scoring of the multi-dimensional payment path evaluation model, and perform path scheduling based on business rules and concurrent request factors.

[0075] Based on the constructed payment path evaluation model, a real-time payment path optimization model is constructed using machine learning technology; training samples are constructed, and supervised learning training samples are constructed based on the real-time scoring of historical transaction data and the evaluation model; for the qth transaction request, the feature vector x is extracted q =(x q1 , x q2 , x q3 , x q4 ), x q1 , x q2 , x q3 , x q4 The characteristic data of amount, currency, time and region information, and the alternative path P = (p q1 , p q2 , ..., p qJ )'s comprehensive utility score (U q1 , U q2 , ..., U qJ ), select the path with the highest utility As the positive sample label y q = 1, other paths are used as negative sample labels y q = 0, forming a training sample (x q , p q ,y q ) ; Randomly divide the training samples into training set Dtrain and test set Dtest.

[0076] Designing a multi-classification machine learning model θ (x, p), where θ is the model parameter to be learned and the input is the transaction feature x iand path set P, the output is the matching score of each path (s 1 ,s 2 , ..., s J ); Considering the heterogeneity of transaction features and path features, a multi-layer perceptron MLP model is used for fusion modeling, and the model structure is designed as follows:

[0077] z path =σ(W path ·P+b path );

[0078] z trans =σ(W trans ·x q +b trans );

[0079] z fuse =σ(W fuse [z path ;z trans ]+b fuse );

[0080] (s 1 ,s 2 , ..., s J )=softmax(W out z fuse +b out );

[0081] Among them, σ(·) is the ReLU activation function, z path represents the hidden layer representation of the payment path feature after being transformed by a layer of neural network; z trans The hidden layer representation of the transaction request feature after being transformed by a layer of neural network; z fuse It represents the fusion representation obtained by concatenating the hidden layer representations of the payment path features and the transaction request features and then transforming them through a layer of neural network; (s 1 ,s 2 , ..., s J ) represents the matching score of each payment path output in the end; W path , W trans , W fuse , W out is the weight matrix to be learned, b path 、b trans 、b fuse 、b out is the bias vector to be learned, [·;·] represents the vector concatenation operation, and softmax(·) normalizes the matching score.

[0082] Based on the training samples (x q , p q,y q )∈Dtrain, the model parameters θ are optimized using the cross entropy loss function and stochastic gradient descent algorithm to minimize the empirical risk:

[0083]

[0084] Where N is the number of training samples, s qj For model f θ (x q , p q )’s j-th output component; the model parameters are updated through the back-propagation algorithm until the loss function converges or the preset number of iterations is reached.

[0085] Based on the test sample (x q , p q ,y q )∈Dtest, evaluate the classification accuracy, precision and recall of the model, perform hyperparameter tuning, and select the model with the best performance For online prediction.

[0086] For real-time transaction requests, extract their feature vectors Generate a set of alternative paths Substitute the trained model and calculate the matching score:

[0087]

[0088] Select the path p with the highest matching degree * = arg max j s j As the optimal path, the request is forwarded to the channel corresponding to the path for processing.

[0089] In order to smooth the concurrent pressure of payment requests and avoid excessive load on individual paths or channels, a path scheduling mechanism is introduced; for each channel c, the historical load level ρ is recorded c (such as request volume, response delay, etc.), when ρ c Exceeding the preset warning line ρ warn When the channel load is high, the transaction requests arriving at the channel are smoothed and valley filled, and are instead assigned to the suboptimal path for processing until the channel load returns to normal levels.

[0090] The S3 step uses machine learning technology to predict the optimal payment path in real time based on multi-dimensional evaluation and scoring, and introduces a load-balancing dynamic path scheduling mechanism; according to the characteristics of the transaction request and the current system status, it intelligently selects the most matching and stable payment path to improve payment success rate and resource utilization.

[0091] S4: Establish payment path anomaly detection and risk control strategies, automatically identify abnormal transaction patterns from massive payment data through machine learning algorithms, and block high-risk transactions in real time.

[0092] Construct abnormal transaction feature engineering to extract feature vectors reflecting transaction abnormalities from payment gateway logs and transaction datasets.

[0093] For the g-th transaction record H g , construct a multidimensional abnormal feature vector H g =(H g1 , H g2 , H g3 , H g4 , H g5 ), where H g1 is the transaction amount feature, which determines whether the transaction amount A exceeds the z times standard deviation of the historical amount distribution of the account, that is, in is the indicator function, μ a and σ a is the mean and standard deviation of account transaction amount; H g2 It is the transaction frequency feature, which determines the number of transactions n of the account within a period of time Δt i Whether it exceeds the q quantile of the historical level during the same period, that is, Where Q q is the quantile function; H g3 : Transaction time characteristics, determine the transaction time t i Whether it is during non-working hours such as late night, weekends, holidays, etc. Right now H g4 It is the counterparty feature, which determines the counterparty account c i Is it a blacklist of high-risk accounts? Right now H g5 The geographical characteristics of the transaction, determine the IP address where the transaction occurred. i Is it located in a high-risk area? Right now

[0094] Constructing abnormal transaction detection model Based on abnormal transaction features, a supervised learning classification model is constructed. The training samples are The label M i ∈0, 1 indicates whether the transaction is abnormal; select the linear discriminant analysis LDA model and learn the discriminant:

[0095]

[0096] Where w is the feature weight vector, b is the threshold bias, sigmoid is the Logistic function, and the discriminant value Mapped to anomaly probability.

[0097] The training objective of the model is to minimize the Iogistic regression loss:

[0098]

[0099] Where λ is the L2 regularization coefficient to prevent overfitting; the gradient descent method is used to learn the model parameters:

[0100]

[0101] Where η is the learning rate; the accuracy, precision, recall and other indicators are evaluated on the test set to perform parameter tuning.

[0102] Deploy the trained anomaly detection model to the payment gateway and extract the anomaly feature vector H from the real-time transaction request. g , input the model for real-time prediction and get the abnormal probability Set the abnormality judgment threshold θ, when When the transaction is considered abnormal, a rejection strategy is implemented to block the transaction request.

[0103] In order to reduce the misjudgment rate and improve the detection rate of high-risk transactions, a hierarchical blocking strategy is designed and the risk threshold ψ is set. 1 and ψ 2 ,0<ψ 1 <ψ 2 <1, based on the predicted value of abnormal probability Divide into three risk levels: low, medium and high, and take corresponding disposal measures according to the risk level:

[0104] like Classify the risk level as low risk and release the transaction;

[0105] like The risk level is classified as medium risk, triggering the manual review process and suspending the transaction;

[0106] like The risk level is classified as high risk, transactions are blocked directly, and restrictive measures are taken on the account.

[0107] Threshold ψ 1 and ψ 2 Flexibly set according to business strategy and risk appetite.

[0108] Through manual review and customer complaint mechanisms, new abnormal transaction annotation data is continuously generated to expand the training data set, and the detection model is regularly retrained and updated so that the model can adapt to the latest risk environment.

[0109] Step S4 uses machine learning algorithms to automatically identify abnormal transaction patterns from massive payment data and build a real-time anomaly detection model. Combined with the risk control strategy of graded blocking, it can promptly detect and prevent high-risk transactions and control payment risks. At the same time, through continuous optimization and iteration, the system can intelligently respond to ever-changing risky transactions.

[0110] Example 2

[0111] Reference Figure 2 , which is the second embodiment of the present invention, provides a cross-border payment path optimization system based on multidimensional data analysis.

[0112] The system includes: a data collection module, a multi-dimensional payment path evaluation module, a payment path optimization module and a payment risk control module.

[0113] The data collection module is used to collect payment path data, collect payment transaction data and exchange rate data by parsing payment gateway logs and calling APIs, and perform data cleaning.

[0114] The multi-dimensional payment path evaluation module is used to construct a multi-dimensional payment path evaluation model, build an evaluation model from the perspectives of payment cost, payment timeliness, payment risk and payment availability, and perform real-time scoring on various indicators of the payment path.

[0115] The payment path optimization module is used to build a payment path optimization model, dynamically select the optimal path based on the real-time scoring of the multi-dimensional payment path evaluation model, and perform path scheduling in combination with business rules and concurrent request factors.

[0116] The payment risk control module is used to establish payment path anomaly detection and risk control strategies, automatically identify abnormal transaction patterns from massive payment data through machine learning algorithms, and block high-risk transactions in real time.

[0117] Example 3

[0118] The present invention also provides an electronic device, comprising: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; the processor executes the cross-border payment path optimization method based on multidimensional data analysis by calling the computer program stored in the memory.

[0119] The present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the cross-border payment path optimization method based on multidimensional data analysis.

[0120] In addition, the parts of the above technical solutions provided in the embodiments of the present invention that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0121] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A cross-border payment path optimization method based on multidimensional data analysis, characterized in that: include: Collect payment path data, collect payment transaction data and exchange rate data by parsing payment gateway logs and calling APIs, and perform data cleaning; Construct a multi-dimensional payment path evaluation model from the perspectives of payment cost, payment timeliness, payment risk, and payment availability, and perform real-time scoring on various payment path indicators; Build a payment path optimization model, dynamically select the optimal path based on the real-time scoring of the multi-dimensional payment path evaluation model, and perform path scheduling based on business rules and concurrent request factors; Establish payment path anomaly detection and risk control strategies, automatically identify abnormal transaction patterns from massive payment data through machine learning algorithms, and block high-risk transactions in real time.

2. The cross-border payment path optimization method based on multidimensional data analysis according to claim 1 is characterized in that: Extract payment log files from the payment gateway of the cross-border payment path. For each log file, extract the payment data in the log file and convert it into structured transaction data; define the data structure of a single transaction record as a tuple; Determine the set of currency pairs for which exchange rates need to be collected. For each currency pair, call the exchange rate API to obtain the exchange rate time series over a period of time; match the exchange rate data with the transaction data, and associate the transaction records with the exchange rate; Clean the original transaction data set, remove duplicate transaction records, and filter by transaction status to eliminate failed, abnormal, and test abnormal transactions; convert the transaction amount into the amount of the base currency according to the exchange rate; sort by time to obtain the transaction time series; generate the cleaned transaction data set; Based on the cleaned transaction data set, aggregate by channel, scenario and time dimension to generate summary statistics of different granularities; define statistical indicators, including total transaction amount, average arrival time and arrival success rate; Calculates the specified payment path statistical indicator value.

3. The cross-border payment path optimization method based on multidimensional data analysis according to claim 2 is characterized in that: For each path in the payment path set, based on the statistical feature vector of the payment path, evaluation indicators are constructed from multiple dimensions including payment cost, payment timeliness, payment risk, and payment availability: The payment cost refers to the transaction cost of the path, including the handling fee rate and exchange loss rate of each channel and node; the payment time refers to the end-to-end processing time of the path, including forwarding delay, queuing waiting time and cross-border settlement time; The payment risk refers to the potential risk of the path, taking into account the risk of fund theft, counterparty default and system failure. The payment availability refers to the technical availability of the path, taking into account the impact of node failure and network interruption. The evaluation indicators are constructed based on multiple dimensions to construct a comprehensive utility score of the path.

4. The cross-border payment path optimization method based on multidimensional data analysis according to claim 3 is characterized in that: Based on the constructed payment path evaluation model, machine learning technology is used to build a real-time payment path optimization model; Construct training samples: Based on historical transaction data and real-time scoring of the evaluation model, construct training samples for supervised learning; Randomly divide the training samples into training set and test set; Design a multi-classification machine learning model with transaction features and path sets as input and matching scores of each path as output; consider the heterogeneity of transaction features and path features and use a multi-layer perceptron (MLP) model for fusion modeling; Based on the training set samples, the model parameters are optimized using the cross entropy loss function and the stochastic gradient descent algorithm, and the model parameters are updated using the back propagation algorithm until the loss function converges or the preset number of iterations is reached; Based on the test set samples, the classification accuracy, precision, and recall of the model are evaluated, hyperparameters are tuned, and the model with the best performance is selected for optimal path prediction.

5. The cross-border payment path optimization method based on multidimensional data analysis according to claim 4 is characterized in that: For transaction requests arriving in real time, extract their feature vectors. Generate a set of candidate paths, substitute them into the trained model, and calculate the matching score; Select the path with the highest matching degree as the optimal path, and forward the request to the channel corresponding to the path for processing; A path scheduling mechanism is introduced to record the historical load level for each channel. When the load level exceeds the preset warning line, the transaction requests arriving at the channel are smoothed and valley-filled, and are instead allocated to the suboptimal path for processing until the channel load returns to normal levels.

6. The cross-border payment path optimization method based on multidimensional data analysis according to claim 5 is characterized in that: Construct abnormal transaction feature engineering to extract feature vectors reflecting transaction abnormalities from payment gateway logs and transaction datasets; Based on the transaction record data, a multi-dimensional abnormal feature vector is constructed; an abnormal transaction detection model is constructed, and based on the abnormal transaction features, a supervised learning classification model is constructed to detect whether the transaction data has abnormal transaction features.

7. The cross-border payment path optimization method based on multidimensional data analysis according to claim 6 is characterized in that: Deploy the trained anomaly detection model to the payment gateway, extract the anomaly feature vector of the real-time transaction request, input it into the model for real-time prediction, and obtain the anomaly probability; Set an abnormality judgment threshold. When the abnormal probability is greater than the abnormality judgment threshold, the transaction is judged as abnormal, and a rejection strategy is implemented to block the transaction request. Design a graded blocking strategy, set multi-level risk thresholds, divide the risk into three levels: low, medium and high according to the abnormal probability prediction value, and take corresponding disposal measures according to the risk level.

8. A cross-border payment path optimization system based on multidimensional data analysis, which is used to implement the cross-border payment path optimization method based on multidimensional data analysis according to any one of claims 1 to 7, characterized in that: include: Data collection module, multi-dimensional payment path evaluation module, payment path optimization module and payment risk control module; The data collection module is used to collect payment path data, collect payment transaction data and exchange rate data by parsing payment gateway logs and calling APIs, and perform data cleaning; The multi-dimensional payment path evaluation module is used to construct a multi-dimensional payment path evaluation model, build an evaluation model from the perspectives of payment cost, payment timeliness, payment risk and payment availability, and perform real-time scoring on various indicators of the payment path; The payment path optimization module is used to build a payment path optimization model, dynamically select the optimal path based on the real-time scoring of the multi-dimensional payment path evaluation model, and perform path scheduling in combination with business rules and concurrent request factors; The payment risk control module is used to establish payment path anomaly detection and risk control strategies, automatically identify abnormal transaction patterns from massive payment data through machine learning algorithms, and block high-risk transactions in real time.

9. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; the processor executes the cross-border payment path optimization method based on multidimensional data analysis as described in any one of claims 1 to 7 by calling the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer executes the cross-border payment path optimization method based on multidimensional data analysis as described in any one of claims 1 to 7.

Images