Quasi-real-time bank account checking management and control method and system

Abstract

The invention discloses a quasi-real-time bank reconciliation management and control method and system, and the method comprises the steps: obtaining a data source, carrying out the collection of data to a data storage platform, transmitting the data in the data storage platform to a real-time flow calculation platform, creating a calculation task according to an account checking detection rule, performing abnormality judgment to obtain a transaction recognition result for storage, positioning a business system and a link in which an abnormal transaction occurs, and intelligently analyzing the type of a technical fault causing the abnormality by using a model constructed by a machine learning algorithm; and for the identified abnormal transaction, carrying out message notification on related personnel. According to the invention, quasi-real-time reconciliation detection of fund transactions is realized, and the timeliness of reconciliation detection is improved; the type of a technical fault causing the abnormal transaction is intelligently analyzed through a machine learning model, so that technicians can conveniently and timely check and repair the fault, and a bank is helped to reduce the fund risk.

Description

Technical Field

[0001] The present invention belongs to the field of informatization feature engineering, and particularly relates to a quasi-real-time bank reconciliation control method and system. Background Art

[0002] With the development and popularization of bank retail business, the transaction volumes of receipts and payments for various businesses such as consumer credit have rapidly increased, and the fund settlement link has become increasingly complex. Completing a business transaction often requires invoking multiple in-house systems, and transaction data is scattered in each link of the invocation link. The longer the transaction link, the greater the probability of accidental accounting-related system failures, including some potential cases of running with problems for a long time. At the same time, problems such as inconsistent units of key elements such as amount and interest rate, interruption of the invocation chain, and short payments caused by idempotency failure may all lead to the occurrence of fund security incidents. Fund security is the bottom line that the banking industry must adhere to.

[0003] Currently, there is little public information on the bank reconciliation control solutions in the banking industry. The fund security prevention and control in industries such as e-commerce and online shopping are in a leading state, but no mature and reusable digital solutions have been introduced yet. Generally speaking, the banking industry usually adopts the method of regular reconciliation for self-checking accounts. Reconciling accounts at regular intervals has poor timeliness. Once problems occur, it is difficult to discover and stop losses in a timely manner, and more manpower is required to handle wrong-account funds and manual adjustment of accounts in the later stage. In addition, for abnormal accounting links, it usually takes a lot of time and effort for technical personnel to troubleshoot faults, and there is a lack of means for fault classification prompts and quick positioning of technical faults. Summary of the Invention

[0004] In view of the above problems, the present invention provides a quasi-real-time bank reconciliation control method and device, which innovatively realizes quasi-real-time reconciliation detection of fund transactions and intelligent classification of technical faults, can improve the timeliness of reconciliation detection, assist technical personnel to carry out troubleshooting work efficiently, and help banks better ensure transaction security.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: a quasi-real-time bank reconciliation control method, which includes the following steps:

[0006] S1: Maintain the data sources and transaction link information of each business system of the bank;

[0007] S2: Obtain the data source, and use a data synchronization tool to collect the data source to a data storage platform. The data source includes transaction data generated by the business system to be reconciled and evaluated;

[0008] S3: Abnormal transaction identification. Use the message middleware to send the data in the data storage platform to the real-time stream computing platform, create a calculation task according to the reconciliation detection rules, and perform abnormal judgment to obtain the transaction identification result. The transaction identification result includes: normal transaction, suspected abnormal transaction;

[0009] S4: Identification result storage and analysis. Store the identification result in the high-speed query engine of the data storage and analysis platform, locate the relevant business systems and links where the abnormal transaction occurs, and use the model constructed by the machine learning algorithm to intelligently analyze the technical fault type that causes the abnormality;

[0010] S5: Message notification for real-time warning. For the identified abnormal transactions, notify the relevant personnel in time in the form of message notification for handling.

[0011] Furthermore, the maintenance elements of the data source in S1 include: data source type, business system table name, meaning and name of data fields; the maintenance elements of the transaction link information include: transaction link name, transaction flow number identification code, and transaction link details.

[0012] Furthermore, the data storage platform is a temporary platform for storing data, specifically a distributed data service platform.

[0013] Furthermore, the reconciliation detection rules include: according to the debit and credit identifiers under the same account and counterparty account in the transaction flow data, judge the transaction situation, calculate and judge the upstream and downstream transaction status, transaction amount, and debit and credit relationship of each transaction flow respectively, and output the reconciliation result according to the judgment result.

[0014] Furthermore, the debit and credit identifiers include debit and credit; the upstream and downstream transaction statuses include True (success), False (failure), and Error (abnormality);

[0015] The calculation and judgment include whether there is a situation of unbalanced accounts and inconsistent upstream and downstream transaction statuses. If so, output the reconciliation result identifier of C002 suspected abnormal transaction; if not, output the reconciliation result identifier of C001 normal transaction.

[0016] Furthermore, during the process of storing the identification result in the high-speed query engine of the data storage and analysis platform, it is also necessary to display the transaction flow information corresponding to the suspected abnormal details, the basic information of the two parties and multiple participants in the transaction, and display the account adjustment plan.

[0017] Furthermore, the types of technical failures that lead to anomalies intelligently analyzed by the model constructed using machine learning algorithms include: using the XGBoost algorithm, obtaining historical abnormal transaction data as training samples, and training a preset initial XGBoost model to obtain a final XGBoost model. Specifically:

[0018] S4.1. Obtain data related to historical abnormal transaction events for a certain type of abnormal scenario. The data features include: business name, relevant system name where the anomaly occurs, call link, upstream and downstream amounts, upstream and downstream transaction statuses, and upstream and downstream transaction times.

[0019] S4.2. Preprocess the data, including handling missing values, handling outliers, and removing redundant features of the upstream and downstream according to feature correlations.

[0020] S4.3. Construct an initial XGBoost model, extract a preset proportion of abnormal transaction data samples as training data, and train the initial XGBoost model to obtain a final XGBoost model.

[0021] Furthermore, S4.3 includes: constructing an initial XGBoost model:

[0022]

[0023] In the formula, is the predicted value output by the model, x i represents the feature vector of the i-th data point, f k is a decision tree, f k (x i ) is the output result of the k-th tree, and F is the set space of decision trees;

[0024] Extract a preset proportion of abnormal transaction data samples as training data. The training dataset is expressed as:

[0025] D i =(x i , y i ): i = 1, 2,..., n; x i ∈R p ; y i ∈R

[0026] n is the number of samples, each sample has p features, and y i is the true value;

[0027] Extract a preset proportion of abnormal transaction data samples as training data, and train the initial XGBoost model to obtain a final XGBoost model;

[0028] During the training process, the objective function is It includes two components. One part is an arbitrarily differentiable loss function used to control the empirical risk of the model; the other part is a regularization term that controls the model complexity used to control the structural risk of the current tree, where T and ω are the number of leaf nodes of the tree and the leaf weight value respectively, γ is the leaf tree penalty coefficient, and λ is the leaf weight penalty coefficient;

[0029] Output of abnormal transaction failure type results; there are a total of n classifications S represented by numerical values k , k ∈ (0, n], where n represents the number of classifications; then the softmax function is:

[0030]

[0031] In the formula, i represents a certain classification in k, and g i represents the value of a certain classification; based on the above softmax function, the final loss function is expressed as:

[0032]

[0033] Among them, P(S c ) is the output value of the softmax function, and y c is the true value of the sample, the smaller it is, the closer it is to the true event result;

[0034] In the process of automatic classification of abnormal transactions, for the input data to be classified, after passing through the machine learning model obtained from the previous training, the final fault recognition result Res is expressed as:

[0035]

[0036] In the formula, the smaller the loss function l, the greater the probability of the corresponding technical fault predicted by softmax, and the technical fault result Res output by automatic recognition is the technical fault type with the highest probability.

[0037] On the other hand, this specification also provides a quasi-real-time bank reconciliation control system, which includes: an information maintenance module for maintaining business system information, and the business system information includes business system data sources, transaction link information, etc.;

[0038] A data acquisition module for acquiring data sources and aggregating the data sources to a distributed data service platform according to a data synchronization tool, and the data sources include transaction data generated by business systems to be reconciled and evaluated;

[0039] Anomaly recognition module, used for anomaly transaction recognition. It uses a message middleware to send data in the distributed data service platform to the real-time stream computing platform, creates calculation tasks according to the reconciliation detection rules, makes anomaly judgments through quasi-real-time reconciliation rules, and obtains anomaly recognition results;

[0040] Storage and analysis module, used for storing and analyzing the recognition results. It stores the recognition results in the high-speed query engine of the data storage and analysis platform, locates the relevant business systems and links where anomalies occur, and combines the recognition results with the systems and links where anomalies occur. Through a machine learning model, it intelligently analyzes the types of technical failures that lead to anomaly transactions;

[0041] Early warning notification module, used for real-time early warning of anomaly events. For the identified and analyzed anomaly events, it notifies relevant personnel for handling to stop losses in a timely manner.

[0042] On the other hand, this specification also provides a quasi-real-time bank reconciliation control device, including a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, it implements a quasi-real-time bank reconciliation control method as described above.

[0043] Beneficial effects:

[0044] (1) Along with the occurrence of transactions, through stream computing technology to process transaction data to achieve quasi-real-time prevention and control of anomaly transactions, and then combined with offline batch account reconciliation, the probability of bank anomaly transactions can be minimized technically. (2) From the perspective of bank accounting bookkeeping, sort out the transaction flows of each business system, string together several flows of a single transaction with the transaction serial number, and abstract the main detection elements such as transaction account, amount, debit / credit identifier, and transaction status, shielding the differences in business scenarios, elements, processes, and details, so that operators do not need to have a deep understanding of the accessed business, and only need to ensure that the key elements of the transaction flow data are not missing. (3) The distributed data service platform collects the transaction data of each business system, which is convenient for subsequent links to access, and can effectively avoid the impact of reconciliation detection, etc. on the production environment. It has no intrusion into the existing business system, the business system is unaware, and it has good scalability. (4) Detect anomaly events during the transaction process based on the reconciliation control rules. After identifying an anomaly, it can accurately display the links and links where the anomaly transaction occurs, and alarm and notify the management personnel to handle it quickly. (5) Store the result data after stream computing in the data service platform, which is convenient for subsequent multi-dimensional analysis of the anomaly transaction system and transaction link, and identify the type of technical failure through an intelligent model to help technical personnel quickly locate the failure. Description of the drawings

[0045] Figure 1 It is a schematic diagram of the reconciliation control process provided by the embodiment of the present invention;

[0046] Figure 2 This is the position of the technical-level quasi-real-time reconciliation control in the architecture of the embodiments of the present invention;

[0047] Figure 3 It is a schematic structural diagram of a bank reconciliation control system;

[0048] Figure 4 It is a schematic structural diagram of a quasi-real-time bank reconciliation control device provided by the embodiments of the present invention. Specific Embodiments

[0049] The following further elaborates on the specific embodiments of the present invention in conjunction with the accompanying drawings.

[0050] As Figure 1 shown, a quasi-real-time bank reconciliation control method provided by the present invention comprises the following specific steps:

[0051] S1: Maintain the information of each business system, where the business system information includes the data source of the business system and transaction link information; the maintenance of the information of each business system includes: data source maintenance, and the elements of the data source maintenance include data source type, business system table name, meanings and names of data fields; transaction link maintenance, and the transaction link maintenance includes transaction link name, transaction serial number identification code, and transaction link details.

[0052] S2: Obtain the data source, and use a data synchronization tool to collect the data source to a distributed data service platform, where the data source includes transaction data generated by the business systems to be reconciled and evaluated;

[0053] The obtaining of the data source includes: collecting the incremental transaction data of each business system to the distributed data service platform according to the data synchronization tool for subsequent reconciliation detection of transaction flows. The distributed data service platform is a temporary platform for storing data, and the incremental transaction data is the transaction data generated by the data source described in S1.

[0054] S3: Identify abnormal transactions, use a message middleware to send the data in the distributed data service platform to a real-time stream computing platform, create a computing task according to the reconciliation detection rules, perform abnormal judgment through the reconciliation rules, and obtain a transaction identification result, where the transaction identification result includes: normal transactions and suspected abnormal transactions. From the perspective of architecture positioning, the position of the technical-level quasi-real-time reconciliation control in the architecture of the embodiments of the present invention is as Figure 2 shown;

[0055] The reconciliation detection rules can be specifically described as follows: Based on the debit and credit indicators under the same account and counterparty account in the transaction flow data, the transaction situation is judged, and the upstream and downstream transaction status, transaction amount, and debit and credit relationship of each transaction flow are calculated and judged respectively. According to the judgment results, the reconciliation results are output. The real-time stream computing platform is an existing platform, including Spark, Flink, etc. Through such platforms, real-time calculation and judgment are carried out to achieve near-real-time prevention and control of abnormal transactions.

[0056] The above debit and credit indicators include debit and credit. The reconciliation result indicators include C001 normal transaction, C002 suspected abnormal transaction (including abnormal situations such as unbalanced accounts, inconsistent upstream and downstream transaction status, etc.). The upstream and downstream transaction status includes True success, False failure, and Error exception.

[0057] S4: Storage and analysis of recognition results. The recognition results are stored in the high-speed query engine of the data storage and analysis platform to locate the relevant business systems and links where abnormal transactions occur, and use the model constructed by machine learning algorithms to intelligently analyze the types of technical failures that lead to abnormalities.

[0058] The storage of the recognition results in S4 also includes: displaying the transaction flow information corresponding to the suspected abnormal details, the basic information of the transaction parties and multiple participating parties, and displaying the account adjustment plan.

[0059] The types of technical failures described in S4 include: call chain interruption (A), duplicate call idempotency failure (B), system logic exception (C), inconsistent amount storage and transmission units (D), and other types (E).

[0060] The model constructed by the machine learning algorithm described in S4 includes, but is not limited to, the XGBoost algorithm. Taking the construction of the model by the XGBoost algorithm as an example, it includes obtaining historical abnormal transaction data as training samples and training the preset initial XGBoost model to obtain the final XGBoost model, including the following steps:

[0061] T1: Obtain data related to historical abnormal transaction events for a certain type of abnormal scenario. The data features include: business name, relevant system name where the abnormality occurs, call link, upstream and downstream amounts, upstream and downstream transaction status, and upstream and downstream transaction times.

[0062] T2: Data preprocessing, including missing value processing, outlier processing, removing redundant features of upstream and downstream according to feature correlation, etc.

[0063] T3: Extract a preset proportion of abnormal transaction data samples as training data and train the initial XGBoost model to obtain the final XGBoost model.

[0064] Step T3 specifically includes:

[0065] Construct a model. XGBoost is a supervised ensemble learning algorithm that sums multiple decision trees, and its formula is as follows:

[0066]

[0067] In formula (1), is the output result of the model, x i represents the feature vector of the i-th data point obtained in Step T1, f k is a decision tree, f k (x i ) is the output result of the k-th tree, and F is the set space of decision trees. Among them, the XGBoost algorithm trains the data

[0068] set can be expressed as:

[0069] D i =(x i , y i ): i = 1, 2,..., n; x i ∈R p ; y i ∈R (2)

[0070] In formula (2), n is the number of samples, and each sample has p features.

[0071] Construct the objective function. In machine learning, the loss function is usually used as the objective function, and the fitting goal is to minimize the "residual" (residual = predicted value - true value). XGBoost adds a regularization term Ω(f k ) to the loss function to form the objective function whose formula is as follows:

[0072]

[0073] In formula (3), y i is the true value, is the predicted value. The objective function has two components. One is an arbitrarily differentiable loss function which controls the empirical risk of the model; the other is the regularization term Ω(f k ) that controls the complexity of the model, which controls the structural risk of the current tree. Among them, the regularization term T and w are the number of leaf nodes and leaf weight values of the tree respectively, γ is the leaf tree penalty coefficient, and λ is the leaf weight penalty coefficient. And the judgment of abnormal transaction fault types can be converted into a multi-classification problem, so softmax can be selected as the loss function.

[0074] Output of abnormal transaction failure type results. Assume there are a total of n classifications S represented by numerical values k , k ∈ (0, n], where n represents the number of classifications. Then the softmax function is as follows:

[0075]

[0076] In Equation (4), i represents a certain classification in k, and g i represents the value of a certain classification. Based on the above softmax function, the final loss function can be expressed as:

[0077]

[0078] In Equation (5), P(S c ) is the output value of the softmax function, and y c is the sample true value, the smaller it is, the closer it is to the true event result. In the process of automatic classification of abnormal transaction failures, for the input data to be classified, after passing through the machine learning model obtained from the aforementioned training, the final failure recognition result Res can be expressed as:

[0079]

[0080] In Equation (6), the smaller the loss function l, the greater the probability of the corresponding technical failure predicted by softmax, and the technical failure result Res output by automatic recognition is the technical failure type with the highest probability.

[0081] S5: Message notification for real-time warning. For the identified abnormal transactions, notify relevant personnel in a timely manner in the form of message notifications for handling.

[0082] The present invention also provides a quasi-real-time bank reconciliation control system, as Figure 3 shown, including: an information maintenance module for maintaining business system information, where the business system information includes business system data sources, transaction link information, etc.;

[0083] a data acquisition module for acquiring data sources, and aggregating the data sources to a distributed data service platform according to a data synchronization tool, where the data sources include transaction data generated by business systems to be reconciled and evaluated;

[0084] an abnormal identification module for abnormal transaction identification, using a message middleware to send the data in the distributed data service platform to a real-time stream computing platform, creating a computing task according to the reconciliation detection rules, and performing abnormal judgment through quasi-real-time reconciliation rules to obtain an abnormal identification result;

[0085] A storage and analysis module for storing and analyzing the recognition results, storing the recognition results in the high-speed query engine of the data storage and analysis platform, locating the relevant business systems and links where anomalies occur, and combining the recognition results with the systems and links where anomalies occur to intelligently analyze the types of technical failures that lead to abnormal transactions through a machine learning model;

[0086] An early warning notification module for real-time warning of abnormal events. For the identified and analyzed abnormal events, it notifies relevant personnel for handling to stop losses in a timely manner.

[0087] Corresponding to the embodiment of a quasi-real-time bank reconciliation control method described above, the present invention also provides an embodiment of a quasi-real-time bank reconciliation control device.

[0088] See Figure 4 , an embodiment of a quasi-real-time bank reconciliation control device provided by the embodiment of the present invention includes a memory and one or more processors. An executable code is stored in the memory. When the processor executes the executable code, it is used to implement a quasi-real-time bank reconciliation control method in the above embodiment.

[0089] The embodiment of the quasi-real-time bank reconciliation control device provided by the present invention can be applied to any device with data processing capabilities. This device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful device, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for operation. From the hardware level, as Figure 4 shown, it is a hardware structure diagram of any device with data processing capabilities where the quasi-real-time bank reconciliation control device provided by the present invention is located. In addition to Figure 4 the shown processor, memory, network interface, and non-volatile memory, the device where the embodiment is located in any device with data processing capabilities usually also includes other hardware according to the actual functions of the device with data processing capabilities, which will not be elaborated here.

[0090] The specific implementation process of the functions and roles of each unit in the above device can be specifically seen in the implementation process of the corresponding steps in the above method, which will not be elaborated here.

[0091] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the descriptions in the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0092] An embodiment of the present invention also provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, a quasi-real-time bank reconciliation control method in the above embodiments is implemented.

[0093] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store the data that has been output or will be output.

[0094] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, a quasi-real-time bank reconciliation control method described above is implemented.

[0095] After considering the specification and practicing the content disclosed herein, those skilled in the art will readily think of other embodiments of the present application. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include the common general knowledge or conventional technical means in the technical field not disclosed in the present application. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present application are pointed out by the claims.

[0096] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit this application. This application is not limited to the exact structures that have been described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is only limited by the appended claims.

Claims

1. A quasi-real-time bank reconciliation control method, characterized in that: The method comprises the following steps: S1: Maintain the data source and transaction link information of each bank's business system; S2: Acquire the data source and use the data synchronization tool to aggregate the data source to the data storage platform. The data source includes the transaction data generated by the business system to be reconciled and evaluated; S3: Abnormal transaction identification: using the message middleware to send the data in the data storage platform to the real-time stream computing platform, creating computing tasks according to the reconciliation detection rules, making abnormal judgments and obtaining transaction identification results, which include normal transactions and suspected abnormal transactions; S4: Storage and analysis of identification results: The identification results are stored in the high-speed query engine of the data storage and analysis platform to locate the relevant business systems and links where abnormal transactions occur, and the model built by the machine learning algorithm is used to intelligently analyze the type of technical failure that caused the abnormality. S5: Real-time warning through message notification. For identified abnormal transactions, relevant personnel will be notified in a timely manner in the form of message notification for processing.

2. A quasi-real-time bank reconciliation control method according to claim 1, characterized in that: The maintenance elements of the data source in S1 include: data source type, business system table name, data field meaning and name; the maintenance elements of the transaction link information include: transaction link name, transaction flow number identification code and transaction link details.

3. A quasi-real-time bank reconciliation control method according to claim 1, characterized in that: The data storage platform is a temporary platform for storing data, specifically a distributed data service platform.

4. A quasi-real-time bank reconciliation control method according to claim 1, characterized in that: The reconciliation detection rules include: judging the transaction status according to the loan identification under the same account and the counterparty account in the transaction flow data, calculating and judging the upstream and downstream transaction status, transaction amount and loan relationship of each flow respectively, and outputting the reconciliation result according to the judgment result.

5. A quasi-real-time bank reconciliation control method according to claim 4, characterized in that: The loan identification includes borrowing and lending; the upstream and downstream transaction status includes True success, False failure, and Error exception; The calculation and judgment include whether there is an imbalance in accounts and inconsistency in upstream and downstream transaction status. If so, the reconciliation result mark C002 of the suspected abnormal transaction is output; if not, the reconciliation result mark C001 of the normal transaction is output.

6. A quasi-real-time bank reconciliation control method according to claim 1, characterized in that: In the process of storing the identification results in the high-speed query engine of the data storage and analysis platform, it is also necessary to display the transaction flow information corresponding to the suspected abnormal details, the basic information of the two parties and multiple parties to the transaction, and the account adjustment plan.

7. A quasi-real-time bank reconciliation control method according to claim 1, characterized in that: The model constructed by using the machine learning algorithm to intelligently analyze the type of technical failure that causes the anomaly includes: using the XGBoost algorithm, obtaining historical abnormal transaction data as training samples, training the preset initial XGBoost model, and obtaining the final XGBoost model, specifically: S4.

1. Obtain relevant data on historical abnormal transaction events for a certain type of abnormal scenario. The data features include: business name, name of the relevant system where the abnormality occurred, call link, upstream and downstream amounts, upstream and downstream transaction status, and upstream and downstream transaction time; S4.

2. Preprocess the data, including missing value processing, outlier processing, and removing redundant features of upstream and downstream according to feature correlation; S4.3 constructs an initial XGBoost model, extracts a preset proportion of abnormal transaction data samples as training data, trains the initial XGBoost model, and obtains the final XGBoost model.

8. A quasi-real-time bank reconciliation control method according to claim 7, characterized in that: The S4.3 includes: Building the initial XGBoost model: In the formula, is the predicted value of the model output, x i represents the feature vector of the i-th data point, f k is a decision tree, f k (x i ) is the output result of the kth tree, and F is the set space of decision trees; A preset proportion of abnormal transaction data samples are extracted as training data. The training data set is expressed as: D i =(x i ,y i ):i=1,2,...,n;x i ∈R p ;y i ∈R n is the number of samples, each sample has p features, y i is the true value; Extracting a preset proportion of abnormal transaction data samples as training data, training the initial XGBoost model, and obtaining a final XGBoost model; The objective function during training is It consists of two components, one of which is an arbitrarily differentiable loss function It is used to control the empirical risk of the model; the other part is the regularization term that controls the complexity of the model Used to control the structural risk of the current tree, where T and w are the number of tree subnodes and leaf weight values, γ is the leaf tree penalty coefficient, and λ is the leaf weight penalty coefficient; Abnormal transaction failure type result output; there are n categories S represented by numerical values k , k∈(0,n], where n represents the number of categories; then the softmax function is: In the formula, i represents a category in k, g i Represents the value of a certain classification; based on the above softmax function, the final loss function is expressed as: Among them, P(S c ) is the output value of the softmax function, yc is the true value of the sample, The smaller it is, the closer it is to the actual event result; In the process of automatic classification of abnormal transaction faults, after the input data to be classified is trained with the machine learning model obtained above, the final fault identification result Res is expressed as: The smaller the loss function is, the greater the probability of the corresponding technical failure predicted by softmax is. The technical failure result Res output by automatic identification is the technical failure type with the highest probability.

9. A quasi-real-time bank reconciliation control system for implementing the method according to any one of claims 1 to 8, characterized in that: The system includes: an information maintenance module for maintaining business system information, wherein the business system information includes business system data source and transaction link information; A data acquisition module is used to acquire a data source and aggregate the data from the data source to a distributed data service platform according to a data synchronization tool. The data source includes transaction data generated by a business system to be reconciled and evaluated; The anomaly identification module is used to identify abnormal transactions. It uses the message middleware to send data from the distributed data service platform to the real-time stream computing platform, creates computing tasks based on the reconciliation detection rules, and uses quasi-real-time reconciliation rules to make anomaly judgments and obtain anomaly identification results. The storage and analysis module is used to store and analyze the recognition results. The recognition results are stored in the high-speed query engine of the data storage and analysis platform, and the relevant business systems and links where the anomaly occurs are located. Combining the recognition results with the systems and links where the anomaly occurs, the type of technical failure that caused the abnormal transaction is intelligently analyzed through the machine learning model; The early warning notification module is used to warn of abnormal events in real time. For abnormal events that have been identified and analyzed, relevant personnel are notified to handle them and stop losses in time.

10. A quasi-real-time bank reconciliation control device, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the processor executes the executable code, a quasi-real-time bank reconciliation control method as described in any one of claims 1 to 8 is implemented.

Images