CN111127201A - Financial anti-money laundering cloud computing resource optimal allocation system and method based on SMDP - Google Patents

Financial anti-money laundering cloud computing resource optimal allocation system and method based on SMDP Download PDF

Info

Publication number
CN111127201A
CN111127201A CN201911224355.8A CN201911224355A CN111127201A CN 111127201 A CN111127201 A CN 111127201A CN 201911224355 A CN201911224355 A CN 201911224355A CN 111127201 A CN111127201 A CN 111127201A
Authority
CN
China
Prior art keywords
suspicious transaction
money laundering
state
transaction report
resource
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201911224355.8A
Other languages
Chinese (zh)
Inventor
梁宏斌
洪鑫涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN201911224355.8A priority Critical patent/CN111127201A/en
Publication of CN111127201A publication Critical patent/CN111127201A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange

Landscapes

  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Engineering & Computer Science (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Technology Law (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)

Abstract

The invention discloses a financial anti-money laundering cloud computing resource optimal allocation system and a financial anti-money laundering cloud computing resource optimal allocation method based on SMDP, wherein after a suspicious transaction report is received by the system, an access control function module consults a risk level of the suspicious transaction report and whether the system has an available anti-money laundering resource unit or not to a system resource management function module; if the suspicious transaction report has available anti-money laundering resource units and the suspicious transaction report is determined to be analyzed immediately, allocating one or more anti-money laundering resource units to analyze and investigate the suspicious transaction report; otherwise the analysis process will be delayed. The system state is defined as the number of the high risk service suspicious transaction reports and the low risk service suspicious transaction reports which are analyzed and investigated in the money laundering prevention system and the number formed by the types of the current events; and the maximum long-term benefit of the anti-money laundering system is deduced, an optimal decision scheme for resource optimal allocation management of the anti-money laundering system is obtained, and the anti-money laundering system can obtain the maximum system benefit.

Description

Financial anti-money laundering cloud computing resource optimal allocation system and method based on SMDP
Technical Field
The invention relates to the technical field, in particular to a financial anti-money laundering cloud computing resource optimal allocation system and method based on SMDP.
Background
With the rapid development of economic globalization and free capital flow, various capital flows with money laundering as the final purpose will cause huge impact on the stability of the macroscopic economy and financial safety of China. Money laundering is an amplifier of economic crimes, and as the number of terrorist violent incidents increases year by year around the world, investigations show that money laundering and terrorist activities have close relationship, and provide fund support for terrorism. Money laundering has been recognized by international society as one of the typical "non-traditional safety problems" after cold warfare, and has seriously threatened many aspects such as social politics, economy, law, public order, and the like. In recent years, China has been struggling to comply with the international anti-money laundering trend in the anti-money laundering field.
Awareness of financial institutions of risk of anti-money laundering and anti-terrorist financing is gradually increasing, and the ability to take anti-money laundering measures and process business is gradually becoming more and more practical. But as an "economiser" financial institution, the primary motivation is to seek maximization of personal interest. In the face of the enhancement of the supervision of China people banks, in order to avoid the risk and possible reputation loss of punished places to the maximum extent, financial institutions select excessive defensive reporting behaviors. Therefore, the number of suspicious reports received by the Chinese anti-money laundering monitoring and analyzing center is increased rapidly, but the marginal information value of the reports is decreased progressively, so that the capacity of a financial information analysis department for discriminating suspected crimes and money laundering activities is hindered. In the report 2013 of the Chinese anti-money laundering report, 2453.10 ten thousand suspicious transaction reports are received by a Chinese anti-money laundering monitoring and analyzing center in 2013, 165 suspected money laundering clues are actively transferred to departments such as the ministry of public security and the like, and a national inspection agency approves and catches criminal cases of suspected money laundering 6. In the field of anti-money laundering, monitoring, analyzing multimedia data and reporting suspicious transactions requires the use of significant anti-money laundering resources (e.g., human resources, computing resources, communication resources, storage resources, etc.), while requiring financial institutions to invest significant compliance costs[4]. With the continuous demand for improvement of anti-money laundering intelligent analysis, identification and monitoring technology and the continuous increase of multimedia data volume, the anti-money laundering cost is increased greatly, so that the anti-money laundering compliance becomes a burden. Years of research by the international pythagorean corporation have shown that the rapidly growing compliance costs of anti-money laundering regulations have become a continuing phenomenon in the development of the international financial industry. The global anti-money laundering survey in 2014 by the international beginner accounting corporation indicates that the anti-money laundering compliance cost of the banking industry continues to increase by 53%, 40% over the prediction in 2011. Therefore, under the condition of increasing anti-money laundering resource consumption and compliance costThe ability to optimally allocate and utilize limited anti-money laundering resources, monitor and analyze multimedia data of suspicious transactions to maximize anti-money laundering benefits is becoming an active research focus for the effective management of anti-money laundering resources (compliance benefits as referred to herein are generally considered to be costs incurred to avoid reputation risks and fines).
Scholars at home and abroad have made a lot of research on economic analysis of money laundering of financial institutions and effectiveness of suspicious transaction reporting systems, and Masciand aro and Filloto describe the link between effectiveness of bank money laundering supervision and features of compliance costs. The important research result is that only when the regulatory agency properly considers the problem of compliance cost, the anti-money laundering participating parties can benefit together. Geiger and Wuensch state that executing anti-money laundering puts a tremendous compliance cost pressure on financial institutions, making the effectiveness and efficiency of financial institutions in the field of anti-money laundering for upstream crimes doubtful. The evidence research of Araujo (2010) leads to the conclusion that the construction of an anti-money laundering related system and the anti-money laundering prevention are more important than the direct identification of the crime of money laundering. The literature reviews that various anomaly detection technologies are widely applied to processing suspicious data, such as data mining, neural networks, genetic algorithms, support vector machines and the like. However, according to the existing research and study, no literature exists in the field of anti-money laundering research at present, and the research can be carried out on the optimized distribution management of the anti-money laundering resources by constructing an anti-money laundering resource distribution system revenue model according to the priority of suspicious transaction reports.
Disclosure of Invention
In view of the above problems, an object of the present invention is to provide a financial anti-money laundering cloud computing resource optimal allocation system and method based on an anti-money laundering resource allocation model of semi-mahalanobis decision process (SMDP), which utilizes currently available anti-money laundering resources to obtain the maximum anti-money laundering system profit or the minimum anti-money laundering cost. The technical scheme is as follows:
a financial anti-money laundering cloud computing resource optimal allocation system based on SMDP comprises a system access control function module, a system resource management function module and an anti-money laundering resource pool; the anti-money laundering resource pool comprises a plurality of anti-money laundering resource units; after receiving a suspicious transaction report, the system access control function module consults the risk level of the suspicious transaction report and whether the system has an available anti-money laundering resource unit or not to the system resource management function module; if available anti-money laundering resource units exist and the suspicious transaction report is determined to be analyzed immediately, the system resource management functional module allocates one or more anti-money laundering resource units to analyze and investigate the suspicious transaction report; otherwise, delaying the analysis processing of the suspicious transaction report; the decision whether to make an immediate analysis is based on: the method comprises the steps of carrying out modeling analysis on the overall profit of the anti-money laundering system based on an anti-money laundering resource allocation model of a semi-Markov decision process, defining the system state as the number of high-risk service suspicious transaction reports and low-risk service suspicious transaction reports which are analyzed and investigated, forming an array with the types of events of the currently generated suspicious transaction reports reaching the anti-money laundering system or the suspicious transaction reports ending the analysis and investigation, and deducing the maximum long-term profit of the system so as to decide the allocation of system resources.
Further, in the anti-money laundering resource allocation model of the semi-mahalanobis decision process, namely, the AMLRAM model:
(1) the system state of the AMLRAM model is expressed as,
Figure BDA0002301740170000021
wherein the content of the first and second substances,
Figure BDA0002301740170000022
e∈{el,eh,ef0 is less than or equal to αlNl+αhNh≤K；
Figure BDA0002301740170000023
Indicating the state of the system, NlIndicates the number of low risk business suspicious transaction reports l under analysis and investigation in the system, NhThe number of the suspicious transaction reports h of the high-risk service which is analyzed and investigated in the system is shown; e denotes an event, elIndicating that the system receives a low from the FIRisk business suspicious transaction report l, ehIndicating that the system receives a suspicious transaction report h, e from FIfIndicating that a suspicious transaction report analyzed and investigated in the system has finished the analysis process and released the anti-money laundering resource unit occupied by it αlRepresenting the number of anti-money laundering resource units allocated for the low risk business suspicious transaction report l αhRepresenting the number of anti-money laundering resource units distributed for the high-risk service suspicious transaction report h; k represents the total number of anti-money laundering resource units in the system;
(2) the AMLRAM model has the following action sets:
Figure BDA0002301740170000031
wherein the content of the first and second substances,
Figure BDA0002301740170000032
it is indicated that the processing is immediate,
Figure BDA0002301740170000033
it is indicated that the processing is delayed,
Figure BDA0002301740170000034
indicating that the analysis survey has ended;
(3) the profit model of the AMLRAM model is as follows:
x(s,a)-τ(s,a)y(s,a)
where x (s, a) represents the total revenue obtained by the system when the current system state is s and action a is selected, and x (s, a) is calculated by:
Figure BDA0002301740170000035
wherein R islAnd RhRespectively is the one-time income brought by the system when the system immediately processes the low risk service suspicious transaction report l and the high risk service suspicious transaction report h;
y (s, a) is the cost per unit time caused by the occupied money laundering resources when the current system state is s and the selected action is a, and the cost per unit time for analyzing and processing the suspicious transaction report is measured by the occupied money laundering resources and is expressed as follows:
Figure BDA0002301740170000036
τ (s, a) is the expected time for the system to transition to the next state when the current state of the system is s and the selected action is a, i.e. the expected time duration between two decision time points, and is expressed as follows:
Figure BDA0002301740170000037
wherein γ ═ λl+λh+Nlμl+Nhμh；λlDenotes λl: l mean rate of arrival of the system, λhRepresenting the mean rate of arrival of the suspicious high-risk service transaction report h at the system;
Figure BDA0002301740170000038
mean service time representing low risk traffic suspicious transaction report/,
Figure BDA0002301740170000039
the mean service time of the suspicious transaction report h of the high-risk service is represented;
(4) using q (j | s, a) to represent the state transition probability that when the current system state is s and the system selects action a, the next system state is j, then:
a) the current system state is
Figure BDA0002301740170000048
If the decision of system selection is "a" is 0, the next possible system state of the system is "j" respectively1＝<Nl,Nh,el>，j2＝<Nl,Nh,eh>，j3＝<Nl-1,Nh,ef>And j4＝<Nl,Nh-1,ef>，Nl≥1NhNot less than 1; the state transition probability q (j | s, a) of the system at this time is:
Figure BDA0002301740170000041
wherein 0 is not more than αlNl+αhNh≤K；
b) The current system state is
Figure BDA0002301740170000042
If the system takes the decision a 1, the next possible system state is j5＝<Nl+1,Nh,el＞，j6＝<Nl+1,Nh,eh＞，j7＝<Nl,Nh,ef>And j8＝<Nl+1,Nh-1,ef>，NhNot less than 1; the state transition probability q (j | s, a) of the system at this time is:
Figure BDA0002301740170000043
c) in the current state of the system is
Figure BDA0002301740170000044
If the system takes the decision a 1, the next possible system state is j9＝<Nl,Nh+1,el>，j10＝<Nl,Nh+1,eh>，j11＝<Nl-1,Nh+1,ef>，NlNot less than 1 and j12＝<Nl,Nh,ef>(ii) a The state transition probability q (j | s, a) of the system at this time is:
Figure BDA0002301740170000045
(5) the expected discounted revenue for the duration τ (s, a) between two decision points is obtained by applying a discounted revenue model of the Markov decision process, as represented by,
Figure BDA0002301740170000046
wherein the content of the first and second substances,
Figure BDA0002301740170000047
representing a desired discount value between two decision points; tau is1α represents a discount factor;
the maximum expected long-term discount yield of the system when the system state is s is obtained by the formula:
Figure BDA0002301740170000051
where upsilon (j) represents the maximum expected discount revenue for the system when the system status is j,
Figure BDA0002301740170000052
finding a finite constant ω such that it satisfies ω ═ λl+λh+K*max(μl,μh)<Infinity, simultaneously order
Figure BDA0002301740170000053
The maximum expected long-term discount yield after normalization is obtained as:
Figure BDA0002301740170000054
wherein
Figure BDA0002301740170000055
At the same time, the normalized state transition probability is,
Figure BDA0002301740170000056
(6) the probability of the suspicious transaction reports of the AMLRAM model being delayed is defined as the ratio of the number of the suspicious transaction reports to be delayed to the total number of the suspicious transaction reports; accordingly, the delay probability P of the ALMRAM model is derivedDelayingComprises the following steps:
Figure BDA0002301740170000057
simultaneously satisfies 0 ≤ αlNl+αhNhK or less, wherein
Figure BDA00023017401700000511
Figure BDA00023017401700000512
Respectively in the system state for the system<Nl,Nh,el>And<Nl,Nh,eh>the corresponding decision to be taken when taking,
Figure BDA00023017401700000513
and
Figure BDA00023017401700000514
and optimizing and distributing the steady-state probability of the management model for the anti-money laundering resource.
An SMDP-based financial anti-money laundering cloud computing resource optimal allocation method comprises the following steps:
step 1: after receiving a suspicious transaction report, judging the suspicious transaction report to be a low-risk service suspicious transaction report l or a high-risk service suspicious transaction report h;
step 2: determining a system state in which a system is currently located, wherein the system state is represented as:
Figure BDA0002301740170000058
wherein the content of the first and second substances,
Figure BDA0002301740170000059
e∈{el,eh,ef0 is less than or equal to αlNl+αhNh≤K；
Figure BDA00023017401700000510
Indicating the state of the system, NlIndicates the number of low risk business suspicious transaction reports l under analysis and investigation in the system, NhThe number of the suspicious transaction reports h of the high-risk service which is analyzed and investigated in the system is shown; e denotes an event, elIndicating that the system receives a low risk traffic suspicious transaction report l, e from FIhIndicating that the system receives a suspicious transaction report h, e from FIfIndicating that a suspicious transaction report analyzed and investigated in the system has finished the analysis process and released the anti-money laundering resource unit occupied by it αlRepresenting the number of anti-money laundering resource units allocated for the low risk business suspicious transaction report l αhRepresenting the number of anti-money laundering resource units distributed for the high-risk service suspicious transaction report h; k represents the total number of anti-money laundering resource units in the system;
and step 3: determining actions that the system can perform, the set of actions being:
Figure BDA0002301740170000061
wherein the content of the first and second substances,
Figure BDA0002301740170000062
it is indicated that the processing is immediate,
Figure BDA0002301740170000063
it is indicated that the processing is delayed,
Figure BDA0002301740170000064
indicating that the analysis survey has ended;
and 4, step 4: calculating the total revenue x (s, a) obtained by the system when the current system state is s and action a is selected:
Figure BDA0002301740170000065
wherein R islAnd RhRespectively is the one-time income brought by the system when the system immediately processes the low risk service suspicious transaction report l and the high risk service suspicious transaction report h;
and 5: calculating the expenditure y (s, a) of the system in unit time caused by the occupied money laundering resources when the current system state is s and the selected action is a:
Figure BDA0002301740170000066
step 6: calculating the expected duration tau (s, a) between two decision time points when the current system state is s and the decision selected by the system is a:
Figure BDA0002301740170000067
wherein γ ═ λl+λh+Nlμl+Nhμh；
And 7: calculating the state transition probability q (j | s, a) that the next system state is j when the current system state is s and the decision selected by the system is a;
a) the current system state is
Figure BDA0002301740170000068
If the decision of system selection is "a" is 0, the next possible system state of the system is "j" respectively1＝<Nl,Nh,el>，j2＝<Nl,Nh,eh>，j3＝<Nl-1,Nh,ef>And j4＝<Nl,Nh-1,ef>，Nl≥1NhNot less than 1; the state transition probability q (j | s, a) of the system at this time is:
Figure BDA0002301740170000071
wherein 0 is not more than αlNl+αhNh≤K；
b) The current system state is
Figure BDA0002301740170000072
If the system takes the decision a 1, the next possible system state is j5＝<Nl+1,Nh,el>，j6＝<Nl+1,Nh,eh>，j7＝<Nl,Nh,ef>And j8＝<Nl+1,Nh-1,ef>，NhNot less than 1; the state transition probability q (j | s, a) of the system at this time is:
Figure BDA0002301740170000073
c) in the current state of the system is
Figure BDA0002301740170000074
If the system takes the decision a 1, the next possible system state is j9＝<Nl,Nh+1,el>，j10＝<Nl,Nh+1,eh>，j11＝<Nl-1,Nh+1,ef>，NiNot less than 1 and j12＝<Nl,Nh,ef>(ii) a The state transition probability q (j | s, a) of the system at this time is:
Figure BDA0002301740170000075
and 8: calculating the maximum expected long-term discount yield:
first, the expected discount yield z (s, a) for the duration τ (s, a) between two decision points is calculated by applying the discount yield model of the markov decision process:
Figure BDA0002301740170000076
wherein the content of the first and second substances,
Figure BDA0002301740170000077
representing a desired discount value between two decision points; tau is1α represents a discount factor;
the maximum expected long-term discount yield of the system when the system state is s is obtained by the formula:
Figure BDA0002301740170000078
where upsilon (j) represents the maximum expected discount revenue for the system when the system status is j,
Figure BDA0002301740170000081
finding a finite constant ω such that it satisfies ω ═ λl+λh+K*max(μl,μh)<Infinity, simultaneously order
Figure BDA0002301740170000082
The maximum expected long-term discount yield after normalization is obtained as:
Figure BDA0002301740170000083
wherein
Figure BDA0002301740170000084
At the same time, the normalized state transition probability is,
Figure BDA0002301740170000085
and step 9: deriving the delay probability P of ALMRAMDelaying：
Figure BDA0002301740170000086
Simultaneously satisfies 0 ≤ αlNl+αhNhK or less, wherein
Figure BDA0002301740170000087
Figure BDA0002301740170000088
In system state separately for AMLRAS<Nl,Nh,el>And<Nl,Nh,eh>the corresponding decision to be taken when taking,
Figure BDA0002301740170000089
and
Figure BDA00023017401700000810
optimizing and distributing the steady-state probability of the management model for the anti-money laundering resource;
step 10: when delay probability PDelayingIf the value is greater than (less than; otherwise, the suspicious transaction report is delayed for analysis processing.
The invention has the beneficial effects that:
the invention analyzes the system income and the system cost of different types of potential suspicious transactions by using a semi-Markov decision process (SMDP), thereby deducing the optimal anti-money laundering resource allocation management strategy of the anti-money laundering system AMLRAS.
The invention provides an optimized distribution model (AMLRAM) of money laundering resources, and obtains the maximum system income of an AMLRAS of the money laundering system by utilizing the model, and the system income considers the system income and expenditure (not only considering the income brought by the AMLRAS completing money laundering monitoring, but also considering the expenditure of money laundering and the expenditure brought by occupying money laundering system resources).
Drawings
FIG. 1 is a diagram illustrating the architecture and processing flow of the anti-money laundering resource optimization management model of the present invention.
FIG. 2 is a graph of the probability of low risk traffic suspicious transaction reports being delayed in processing at different high risk traffic suspicious transaction reports mean arrival rates.
FIG. 3 is a graph illustrating the probability of high risk traffic suspicious transaction reports being delayed in processing at different high risk traffic suspicious transaction reports mean arrival rates.
FIG. 4 is a graph of the probability of low risk traffic suspicious transaction reports being delayed in processing at different processing mean rates of the suspicious transaction reports.
FIG. 5 is a graph of the probability of high risk traffic suspicious transaction reports being delayed in processing at different processing mean rates of the suspicious transaction reports.
Detailed Description
The invention is described in further detail below with reference to the figures and specific embodiments. The invention mainly focuses on the optimized allocation management of the anti-money laundering resources for analyzing and screening the multimedia data of potential suspicious transactions with different risk levels. The invention adopts an Anti-Money Laundering Resource allocation system (AMLRAS) to identify, evaluate and report potential suspicious transactions, and particularly needs to intensively discriminate and evaluate financial products, services, customers, entities, transactions, geographical positions and the like with high risks. If a transaction shows potential suspicion, a suspicious transaction report is immediately submitted to the AMLRAS, which then decides to allocate money laundering resources immediately or later for further analysis and investigation of the suspicious transaction based on the overall maximum revenue of the system and the risk level of the suspicious transaction report.
In order to analyze the above problem of resource optimization allocation management related to potential suspicious transaction reports, the invention provides an Anti-Money Laundering resource allocation Model (AMLRAM) based on semi-Markov decision process (SMDP), and based on the Model, the invention utilizes the available Anti-Money Laundering resources to obtain the maximum Anti-Money Laundering system benefit orMinimal anti-money laundering costs. Although the Markov model is in many fields (e.g., the field of wireless communications)[14]Field of cloud computing[15,16,17]And the field of smart power grids[18,19]) Has been widely used, but based on the existing research, the SMDP has not been used in the optimization distribution of anti-money laundering resources.
(1) System architecture
Fig. 1 illustrates the structure and processing flow of an Anti-Money laundering resource optimization Allocation Model (AMLRAM) proposed by the present invention. When a suspicious transaction report is submitted to the AMLRAS, the access control function of the system consults the system Anti-Money Laundering Resource management function (Anti-Money clearing Resource allocation Unit, AMLRAU) with the risk level of the suspicious transaction report and the Anti-Money Laundering resources available to the system. The amllau is the smallest anti-money laundering resource unit for analyzing and investigating suspicious transaction reports, including personnel, computing resources, communication resources, storage resources, and the like in the anti-money laundering resources. If the AMLRAS has available AMLRAU resources, and the AMLRAS decides to immediately analyze the suspicious transaction report, the system resource management function module allocates one or more AMLRAUs to analyze and investigate the suspicious transaction report; otherwise, the AMLRAS will delay the suspicious transaction report from the analysis process.
As shown in fig. 1, based on a risk analysis method, the present invention considers two types of potentially suspicious transactions: (1) high-Risk business suspicious transactions (HRO), including high-Risk financial products, services, customers, entities, transactions, and geographic locations; (2) low Risk traffic suspicious transaction (Lower-rice Operation LRO). A high-risk business suspicious transaction indicates that the transaction is at a higher risk than a low-risk business suspicious transaction, more likely to result in FI increased costs, financial loss, reputation loss, and more likely participation in money laundering activities or illegal transactions through FI. Therefore, suspect transaction reports at the HRO level must be immediately prioritized. In order to maximize the overall system yield of money laundering and the number of suspicious transactions processed, different money laundering resources (AMLRAU) are allocated to the suspicious transaction reports of different risk levels for further analysis and investigation. When a suspicious transaction report submitted by the FI to the AMLRAS is decided to be immediately analyzed for processing, the AMLRAS allocates one or more AMLRAS to the HRO or LRO's suspicious transaction report for immediate analysis for processing of the suspicious transaction report.
(2) And (3) system model:
in order to improve the system yield or reduce the system cost of an anti-money laundering resource allocation system (AMLRAS), the present invention divides potential suspicious transactions into two categories, high risk traffic suspicious transactions (HRO) and low risk traffic suspicious transactions (LRO). High-risk business suspicious transactions have a higher risk of money laundering than low-risk business suspicious transactions, resulting in loss of economy and reputation of FI, while increasing costs. Therefore, the suspicious transaction of the high-risk business needs to allocate more money laundering resources for analysis processing. In the following section, for convenience of discussion, the present invention uses l and h to represent low risk traffic suspicious transaction reports and high risk traffic suspicious transaction reports, respectively.
In the model, the resources of the money laundering resource allocation system (AMLRAS) are divided into K parts in total, each part represents one money laundering unit resource (AMLRAU). in the AMLRAS, the money laundering resource management function module of the system allocates α the suspected transaction report of the potential suspected high-risk service and the suspected transaction report of the low-risk service respectivelylAn anti-money laundering resource (AMLRAU) and αhAn anti-money laundering resource (AMLRAU), 0<αl+αh<K. Because money laundering resources are limited (i.e., the number of AMLRAUs is limited), it is an increasingly urgent issue for money laundering systems to balance the relationship between revenue generated by the administration of suspicious transactions by the AMLRAS and expenses generated by the operation of the money laundering system, and to obtain the maximum overall benefit of the money laundering system by optimally allocating money laundering resources. In order to obtain the maximum overall system revenue of the AMLRAS, the AMLRAS needs to decide whether to immediately analyze or delay the analysis and process the potentially suspicious transaction reports (high risk business suspicious transaction reports-HRO and HRO) according to the status of the anti-money laundering resources currently available to the system and the arrival rate of the potentially suspicious transaction reports submitted to the AMLRAS systemLow risk traffic suspicious transaction report-LRO).
In the AMLRAM model, the mean rates of the low-risk service suspicious transaction report l and the high-risk service suspicious transaction report h reaching the AMLRAS are subject to Poisson distribution, and the mean rates are lambdalAnd λh. The analysis investigation time required by the low risk business available transaction report l and the high risk business suspicious transaction report h in the AMLRAS obeys the exponential distribution, and the mean service time is respectively
Figure BDA0002301740170000101
And
Figure BDA0002301740170000102
(3) system state
A request to low risk traffic suspicious transaction report l or high risk traffic suspicious transaction report h of the AMLRAS is considered to be an incoming event; for a low risk business suspicious transaction report l or a high risk business suspicious transaction report h under analysis investigation in the AMLRAS, when it finishes the analysis process and releases the money laundering resources, it can be regarded as an outgoing event. In the present system model, therefore, three types of events are defined,
1) AMLRAS receives a report l of low risk traffic suspicious transaction from FI, using elTo represent;
2) AMLRAS receives a suspicious transaction report h of high-risk traffic from FI, using ehTo represent;
3) when a suspicious transaction report (low risk service suspicious transaction report or high risk service suspicious transaction report) analyzed and investigated in AMLRAS has finished analyzing and processing and released the anti-money laundering resource (AMLRAU) occupied by the suspicious transaction report, the event uses efTo indicate.
In addition, the number of low risk service suspicious transaction reports l and the number of high risk service suspicious transaction reports h which are analyzed and investigated in AMLRAS are respectively NlAnd NhTo indicate. Therefore, the system state of the model can be as followsAs indicated by the general representation of the,
Figure BDA0002301740170000111
wherein the content of the first and second substances,
Figure BDA0002301740170000112
e∈{el,eh,ef0 is less than or equal to αlNl+αhNh≤K。
(4) Action set
In the model, when the system is in the state
Figure BDA0002301740170000118
And receives a report of the suspicious transaction (e)lOr eh) The AMLRAS may choose two actions: immediate processing or delayed processing, respectively denoted as
Figure BDA0002301740170000113
And
Figure BDA0002301740170000114
when the analysis investigation of one suspicious transaction report (low-risk business suspicious transaction report l and high-risk business suspicious transaction report h) has ended and the occupied anti-money laundering resource is released, the action at this time can be expressed as
Figure BDA0002301740170000115
Therefore, to sum up, all the actions of the present model are,
Figure BDA0002301740170000116
(5) profit model
The net income of the AMLRAS system consists of income brought by AMLRAS supervision of suspicious transactions and expenditure brought by analysis and investigation of anti-money laundering resources occupied by suspicious transaction reports. The calculation formula thereof can be expressed as,
x(s,a)-τ(s,a)y(s,a) (1)
where x (s, a) is the total revenue obtained by the AMLRAS system when the current system state is s and action a is selected, y (s, a) is the cost per unit time due to the occupied anti-money laundering resource when the AMLRAS system selects action a when the current system state is s, and τ (s, a) is the expected time for the system to transition to the next state when the AMLRAS system selects action a when the current state is s.
x (s, a) can be calculated from the following formula,
Figure BDA0002301740170000117
wherein R islAnd RhThe system is respectively the disposable income brought when the AMLRAS system immediately processes the low risk service suspicious transaction report l and the high risk service suspicious transaction report h. The cost per unit time y (s, a) for analyzing and processing the suspicious transaction report can be measured by the occupied anti-money laundering resources, which is expressed by the following formula,
Figure BDA0002301740170000121
(6) modeling based on semi-Markov decision system
Typically, a semi-Markov decision process contains six elements[20,21]:1) a system state; 2) an action set; 3) a set of events; 4) deciding a time point; 5) a state transition probability; 6) and (6) earning. The decision time point refers to the time when any event occurs, for example, the time when a low risk service suspicious transaction report l or a high risk service suspicious transaction report h is sent to the AMLRAS by FI, or the time when a suspicious transaction report (low risk service suspicious transaction report l or high risk service suspicious transaction report h) completes analysis investigation in the AMLRAS and releases the money laundering resources (amlraa) occupied by the suspicious transaction report. While the time between two decision time points is usually subject to an exponential distribution, the expected duration between two decision time points when the current system state is s and the system selects decision a is denoted by τ (s, a).
Thus, it is possible to obtain a compound having τ (s, a) of,
Figure BDA0002301740170000122
wherein γ ═ λl+λh+Nlμl+Nhμh。
The state transition probability of the next system state being j when the AMLRAS system selects action a at the current system state being s is represented by q (j | s, a). In the current system state of
Figure BDA0002301740170000123
When (therein)
Figure BDA0002301740170000124
e∈{el,eh,efH), if the decision of the AMLRAS system selection is a-0, then the next possible system state is j, respectively1＝<Nl,Nh,el>，j2＝<Nl,Nh,eh>，j3＝<Nl-1,Nh,ef>(NlNot less than 1) and j4＝<Nl,Nh-1,ef>(Nh≥1)。
Thus, the state transition probability q (j | s, a) is obtained,
Figure BDA0002301740170000125
wherein 0 is not more than αlNl+αhNh≤K。
In the current system state of
Figure BDA0002301740170000126
If the decision taken by the AMLRAS system is that a is 1, then the next possible state of the system is j5＝<Nl+1,Nh,el>，j6＝<Nl+1,Nh,eh>，j7＝<Nl,Nh,ef>And j8＝<Nl+1,Nh-1,ef>(NhNot less than 1). In this case, the state transition probability q (j | s, a) of the AMLRAS system may be expressed as,
Figure BDA0002301740170000131
similarly, the current state of the system is
Figure BDA0002301740170000132
If the decision taken by the AMLRAS system is that a is 1, then the next possible state of the system is j9＝<Nl,Nh+1,el>，j10＝<Nl,Nh+1,eh>，j11＝<Nl-1,Nh+1,ef>(NlNot less than 1) and j12＝<Nl,Nh,ef>. In this case, the state transition probability q (j | s, a) of the AMLRAS system is expressed as,
Figure BDA0002301740170000133
discount revenue model through application of Markov decision process[20,21]The expected discount yield for the duration τ (s, a) between two decision points can be obtained, as expressed by,
Figure BDA0002301740170000134
wherein x (s, a) and y (s, a) are defined in equations (2) and (3), respectively. Thus, the maximum expected long-term discount yield of the AMLRAS system may be obtained as,
Figure BDA0002301740170000135
wherein the content of the first and second substances,
Figure BDA0002301740170000136
a finite constant ω can be found to satisfy ω ═ λl+λh+K*max(μl,μh)<Infinity, simultaneously order
Figure BDA0002301740170000137
Thus, the maximum expected long-term discount yield after normalization can be obtained as:
Figure BDA0002301740170000138
wherein
Figure BDA0002301740170000139
At the same time, the normalized state transition probability is,
Figure BDA00023017401700001310
the probability of delayed processing of suspicious transaction reports of the money laundering resource optimization allocation management model is defined as the ratio of the number of suspicious transaction reports (including low risk business suspicious transaction reports and high risk business suspicious transaction reports) that are delayed processing to the number of total suspicious transaction reports (including low risk business suspicious transaction reports and high risk business suspicious transaction reports). Accordingly, the delay probability ratio P of ALMRAM can be derivedDelayingIn order to realize the purpose,
Figure BDA0002301740170000141
simultaneously satisfies 0 ≤ αlNl+αhNhK or less, wherein
Figure BDA0002301740170000143
Figure BDA0002301740170000144
In system state separately for AMLRAS<Nl,Nh,el>And<Nl,Nh,eh>the corresponding decision to be taken when taking,
Figure BDA0002301740170000145
and
Figure BDA0002301740170000146
and optimizing and distributing the steady-state probability of the management model for the anti-money laundering resource.
In order to construct an efficient AMLRAS anti-money laundering resource optimal allocation management model for suspicious transaction reports, the anti-money laundering resource optimal allocation management model is provided based on the overall AMLRAS system revenue (including income obtained through anti-money laundering, anti-money laundering cost and cost of occupying anti-money laundering resources). The systematic revenue of anti-money laundering can be generally divided into the following[7]:1) intangible benefit: the reputation of compliance departments is largely dependent on the relationship with customers and competitors and the degree of acceptance outside. Reputation risk costs include: direct cost (loss of revenue); indirect costs (customer departure and legal fees that may be incurred) and opportunity costs (giving up other business opportunities); 2) tangible benefits: gains made by avoiding the regulatory punishment of non-compliance. The anti-money laundering cost is typically a compliance cost, and includes the following 4 cost components[4]:1) opportunity/reputation cost; 2) operating costs; 3) the total cost; 4) marginal cost.
Therefore, the system yield of the anti-money laundering resource optimization allocation model of the AMLRAS provided by the invention can be obtained by calculating the following four factors through a system resource management function module: 1) arrival rate and departure rate of high risk business suspicious transaction reports and low risk business suspicious transaction reports; 2) the number of high risk traffic suspicious transaction reports and low risk traffic suspicious transaction reports being analyzed and investigated in the AMLRAS; 3) available anti-money laundering resources (expressed as number of AMLRAUs); 4) and (4) revenue obtained by analyzing and surveying the high-risk business suspicious transaction report and the low-risk business suspicious transaction report. High risk business suspicious transactions have a higher money laundering risk and are more likely to expose financial institutions to economic losses, relative to low risk business suspicious transactions,Increased costs and risk of loss of reputation. Therefore, to ensure risk reduction, analysis investigation of high risk business suspicious transaction reports typically requires more money laundering resources (AMLRAUs) than low risk suspicious transaction reports, and analysis, investigation and identification of high risk business suspicious transaction reports also results in more system revenue for the AMLRAS. Therefore, the anti-money laundering resource optimization allocation management model (AMLRAM) provided by the invention can be used for calculating the maximum expected long-term discount income of the system in the formula (10) under the condition that the anti-money laundering resources are limited
Figure BDA0002301740170000142
The optimal allocation management scheme of the money laundering resources is obtained, and the maximum system benefit or the minimum system expenditure cost of the money laundering system AMLRAS is obtained.
(7) Model performance assessment
The performance of the anti-money laundering resource allocation model (AMLRAM) proposed by the present invention was evaluated by an analog simulation program implemented with Matlab. First, an anti-money laundering resource allocation management system (AMLRAS) is established, and the anti-money laundering resource is calculated by AMLRAU, and is at most 15 and at least 2. Secondly, the mean rates of the low risk service suspicious transaction report l and the high risk service suspicious transaction report h reaching the AMLRAS are 5 and 2 respectively in unit time. If not otherwise stated, the duration of analysis investigation of both the low risk business suspicious transaction report and the high risk business suspicious transaction report in the AMLRAS is 1/6 units of time, namely, mul＝μhThirdly, when the AMLRAS immediately processes a high risk traffic suspicious transaction report, the gained is 0.6, and 2 AMLRAS are allocated to the high risk traffic suspicious transaction report, and similarly, when the AMLRAS immediately processes a low risk traffic suspicious transaction report, the gained is 0.3, and 1 AMLRAS are allocated to the low risk traffic suspicious transaction report, and finally, in order to ensure the convergence of the expected profit calculation, the value of the discount factor α is set to 0.1, that is, α is 0.1.
Fig. 2 shows the probability of delayed processing of a low-risk traffic suspicious transaction report under conditions of different mean-of-arrival rates of high-risk traffic suspicious transaction reports, while fig. 3 shows the probability of delayed processing of a high-risk traffic suspicious transaction report under conditions of different mean-of-arrival rates of high-risk traffic suspicious transaction reports. In fig. 2 and 3, simul represents a simulation value and anal represents a theoretical derivation value. It can be seen from the figure that when there are more anti-money laundering resources in the AMLRAS system (in terms of the number of AMLRAUs), the probability of delayed processing of suspicious transaction reports by the system is lower. In addition, it can be further known from the figure that since the analysis investigation of the high risk business suspicious transaction report requires 2 times of money laundering resources compared with the analysis investigation of the low risk business suspicious transaction report, the high risk business suspicious transaction report is more easily delayed by the AMLRAS system to be processed, especially when the money laundering resources of the AMLRAS system are limited, for example, only 2 available AMLRAS in the AMLRAS system remain. This also explains why in both figures, the probability of high-risk traffic suspicious transaction reports being processed with delay is higher than the probability of low-risk traffic suspicious transaction reports being processed with delay. When the mean rate of the suspicious transaction reports of high-risk services reaching the AMLRAS system is further increased from 2 per unit time to 5, under the condition that money laundering resources of the AMLRAS system are fixed, the probability of delayed processing of the suspicious transaction reports of high-risk services or the probability of delayed processing of the suspicious transaction reports of low-risk services is increased along with the increase of the mean rate of the suspicious transaction reports of high-risk services reaching the AMLRAS system.
Fig. 4 and fig. 5 respectively show the probability of delayed processing of the low risk service suspicious transaction report and the high risk service suspicious transaction report under different processing mean rates of the low risk service suspicious transaction report and the high risk service suspicious transaction report (i.e. the mean rate of the process of analyzing and investigating the high risk service suspicious transaction report and the low risk service suspicious transaction report in the AMLRAS system is finished and releasing the money laundering resources). As can be seen from the figure, the longer the low-risk service suspicious transaction reports and the high-risk service suspicious transaction reports are analyzed and processed in the AMLRAS system, that is, the longer each suspicious transaction report occupies the anti-money-laundering resource in the AMLRAS (that is, the lower the mean rate of releasing the anti-money-laundering resource), the more overhead the AMLRAS system has on each suspicious transaction report, and therefore, the available anti-money-laundering resource is reduced, and the system revenue of the AMLRAS is also reduced. Therefore, in this case, a suspicious transaction report to the AMLRAS system is likely to be rejected.
On the other hand, it can be seen from the figure that as the time for the low risk business suspicious transaction report and the high risk business suspicious transaction report to occupy the money laundering resource decreases, the probability that both the low risk business suspicious transaction report and the high risk business suspicious transaction report are delayed to be processed decreases gradually. This also verifies the performance of the anti-money laundering resource optimization allocation management model proposed by the present invention from another aspect.

Claims (3)

1. A financial anti-money laundering cloud computing resource optimal allocation system based on SMDP is characterized by comprising a system access control function module, a system resource management function module and an anti-money laundering resource pool; the anti-money laundering resource pool comprises a plurality of anti-money laundering resource units; after receiving a suspicious transaction report, the system access control function module consults the risk level of the suspicious transaction report and whether the system has an available anti-money laundering resource unit or not to the system resource management function module; if available anti-money laundering resource units exist and the suspicious transaction report is determined to be analyzed immediately, the system resource management functional module allocates one or more anti-money laundering resource units to analyze and investigate the suspicious transaction report; otherwise, delaying the analysis processing of the suspicious transaction report; the decision whether to make an immediate analysis is based on: the method comprises the steps of carrying out modeling analysis on the overall profit of the anti-money laundering system based on an anti-money laundering resource allocation model of a semi-Markov decision process, defining the system state as the number of high-risk service suspicious transaction reports and low-risk service suspicious transaction reports which are analyzed and investigated, forming an array with the types of events of the currently generated suspicious transaction reports reaching the anti-money laundering system or the suspicious transaction reports ending the analysis and investigation, and deducing the maximum long-term profit of the system so as to decide the allocation of system resources.
2. The SMDP-based financial anti-money laundering cloud computing resource optimal allocation system according to claim 1, wherein in the semi-mahalanobis decision process anti-money laundering resource allocation model, namely, the AMLRAM model:
(1) the system state of the AMLRAM model is expressed as,
Figure FDA0002301740160000011
wherein the content of the first and second substances,
Figure FDA0002301740160000012
e∈{el,eh,ef0 is less than or equal to αlNl+αhNh≤K；
Figure FDA0002301740160000013
Indicating the state of the system, NlIndicates the number of low risk business suspicious transaction reports l under analysis and investigation in the system, NhThe number of the suspicious transaction reports h of the high-risk service which is analyzed and investigated in the system is shown; e denotes an event, elIndicating that the system receives a low risk traffic suspicious transaction report l, e from FIhIndicating that the system receives a suspicious transaction report h, e from FIfIndicating that a suspicious transaction report analyzed and investigated in the system has finished the analysis process and released the anti-money laundering resource unit occupied by it αlRepresenting the number of anti-money laundering resource units allocated for the low risk business suspicious transaction report l αhRepresenting the number of anti-money laundering resource units distributed for the high-risk service suspicious transaction report h; k represents the total number of anti-money laundering resource units in the system;
(2) the AMLRAM model has the following action sets:
Figure FDA0002301740160000014
wherein the content of the first and second substances,
Figure FDA0002301740160000015
it is indicated that the processing is immediate,
Figure FDA0002301740160000016
it is indicated that the processing is delayed,
Figure FDA0002301740160000017
indicating that the analysis survey has ended;
(3) the profit model of the AMLRAM model is as follows:
x(s,a)-τ(s,a)y(s,a)
where x (s, a) represents the total revenue obtained by the system when the current system state is s and action a is selected, and x (s, a) is calculated by:
Figure FDA0002301740160000021
wherein R islAnd RhRespectively is the one-time income brought by the system when the system immediately processes the low risk service suspicious transaction report l and the high risk service suspicious transaction report h;
y (s, a) is the cost per unit time caused by the occupied money laundering resources when the current system state is s and the selected action is a, and the cost per unit time for analyzing and processing the suspicious transaction report is measured by the occupied money laundering resources and is expressed as follows:
Figure FDA0002301740160000022
τ (s, a) is the expected time for the system to transition to the next state when the current state of the system is s and the selected action is a, i.e. the expected time duration between two decision time points, and is expressed as follows:
Figure FDA0002301740160000023
wherein γ ═ λl+λh+Nlμl+Nhμh；λlDenotes λl: l mean rate of arrival of the system, λhRepresenting the mean rate of arrival of the suspicious high-risk service transaction report h at the system;
Figure FDA0002301740160000024
mean service time representing low risk traffic suspicious transaction report/,
Figure FDA0002301740160000025
the mean service time of the suspicious transaction report h of the high-risk service is represented;
(4) using q (j | s, a) to represent the state transition probability that when the current system state is s and the system selects action a, the next system state is j, then:
a) the current system state is
Figure FDA0002301740160000026
If the decision of system selection is "a" is 0, the next possible system state of the system is "j" respectively1＝<Nl,Nh,el>，j2＝<Nl,Nh,eh>，j3＝<Nl-1,Nh,ef>And j4＝<Nl,Nh-1,ef>，Nl≥1 NhNot less than 1; the state transition probability q (j | s, a) of the system at this time is:
Figure FDA0002301740160000031
wherein 0 is not more than αlNl+αhNh≤K；
b) The current system state is
Figure FDA0002301740160000032
If the decision taken by the system is that a is 1,the next possible system state is j5＝<Nl+1,Nh,el>，j6＝<Nl+1,Nh,eh>，j7＝<Nl,Nh,ef>And j8＝<Nl+1,Nh-1,ef>，NhNot less than 1; the state transition probability q (j | s, a) of the system at this time is:
Figure FDA0002301740160000033
c) in the current state of the system is
Figure FDA0002301740160000034
If the system takes the decision a 1, the next possible system state is j9＝<Nl,Nh+1,el>，j10＝<Nl,Nh+1,eh>，j11＝<Nl-1,Nh+1,ef>，NlNot less than 1 and j12＝<Nl,Nh,ef>(ii) a The state transition probability q (j | s, a) of the system at this time is:
Figure FDA0002301740160000035
(5) the expected discounted revenue for the duration τ (s, a) between two decision points is obtained by applying a discounted revenue model of the Markov decision process, as represented by,
Figure FDA0002301740160000036
wherein the content of the first and second substances,
Figure FDA0002301740160000037
representing a desired discount value between two decision points; tau is1α represents a discount factor;
the maximum expected long-term discount yield of the system when the system state is s is obtained by the formula:
Figure FDA0002301740160000038
where upsilon (j) represents the maximum expected discount revenue for the system when the system status is j,
Figure FDA0002301740160000039
finding a finite constant ω such that it satisfies ω ═ λl+λh+K*max(μl,μh)<Infinity, simultaneously order
Figure FDA00023017401600000310
The maximum expected long-term discount yield after normalization is obtained as:
Figure FDA0002301740160000041
wherein
Figure FDA0002301740160000042
At the same time, the normalized state transition probability is,
Figure FDA0002301740160000043
(6) the probability of the suspicious transaction reports of the AMLRAM model being delayed is defined as the ratio of the number of the suspicious transaction reports to be delayed to the total number of the suspicious transaction reports; accordingly, the delay probability P of the ALMRAM model is derivedDelayingComprises the following steps:
Figure FDA0002301740160000044
simultaneously satisfies 0 ≤ αlNl+αhNhK or less, wherein
Figure FDA0002301740160000045
Figure FDA0002301740160000046
Respectively in the system state for the system<Nl,Nh,el>And<Nl,Nh,eh>the corresponding decision to be taken when taking,
Figure FDA0002301740160000047
and
Figure FDA0002301740160000048
and optimizing and distributing the steady-state probability of the management model for the anti-money laundering resource.
3. An SMDP-based optimal allocation method for financial anti-money laundering cloud computing resources is characterized by comprising the following steps:
step 1: after receiving a suspicious transaction report, judging the suspicious transaction report to be a low-risk service suspicious transaction report l or a high-risk service suspicious transaction report h;
step 2: determining a system state in which a system is currently located, wherein the system state is represented as:
Figure FDA0002301740160000049
wherein the content of the first and second substances,
Figure FDA00023017401600000410
e∈{el,eh,ef0 is less than or equal to αlNl+αhNh≤K；
Figure FDA00023017401600000411
Indicating the state of the system, NlIndicates the number of low risk business suspicious transaction reports l under analysis and investigation in the system, NhThe number of the suspicious transaction reports h of the high-risk service which is analyzed and investigated in the system is shown; e denotes an event, elIndicating that the system receives a low risk traffic suspicious transaction report l, e from FIhIndicating that the system receives a suspicious transaction report h, e from FIfIndicating that a suspicious transaction report analyzed and investigated in the system has finished the analysis process and released the anti-money laundering resource unit occupied by it αlRepresenting the number of anti-money laundering resource units allocated for the low risk business suspicious transaction report l αhRepresenting the number of anti-money laundering resource units distributed for the high-risk service suspicious transaction report h; k represents the total number of anti-money laundering resource units in the system;
and step 3: determining actions that the system can perform, the set of actions being:
Figure FDA0002301740160000051
wherein the content of the first and second substances,
Figure FDA0002301740160000052
it is indicated that the processing is immediate,
Figure FDA0002301740160000053
it is indicated that the processing is delayed,
Figure FDA0002301740160000054
indicating that the analysis survey has ended;
and 4, step 4: calculating the total revenue x (s, a) obtained by the system when the current system state is s and action a is selected:
Figure FDA0002301740160000055
wherein R islAnd RhRespectively is the one-time income brought by the system when the system immediately processes the low risk service suspicious transaction report l and the high risk service suspicious transaction report h;
and 5: calculating the expenditure y (s, a) of the system in unit time caused by the occupied money laundering resources when the current system state is s and the selected action is a:
Figure FDA0002301740160000056
step 6: calculating the expected duration tau (s, a) between two decision time points when the current system state is s and the decision selected by the system is a:
Figure FDA0002301740160000057
wherein γ ═ λl+λh+Nlμl+Nhμh；
And 7: calculating the state transition probability q (j | s, a) that the next system state is j when the current system state is s and the decision selected by the system is a;
a) the current system state is
Figure FDA0002301740160000058
If the decision of system selection is "a" is 0, the next possible system state of the system is "j" respectively1＝<Nl,Nh,el>，j2＝<Nl,Nh,eh>，j3＝<Nl-1,Nh,ef>And j4＝<Nl,Nh-1,ef>，Nl≥1 NhNot less than 1; the state transition probability q (j | s, a) of the system at this time is:
Figure FDA0002301740160000059
wherein 0 is not more than αlNl+αhNh≤K；
b) The current system state is
Figure FDA0002301740160000061
If the system takes the decision a 1, the next possible system state is j5＝<Nl+1,Nh,el>，j6＝<Nl+1,Nh,eh>，j7＝<Nl,Nh,ef>And j8＝<Nl+1,Nh-1,ef>，NhNot less than 1; the state transition probability q (j | s, a) of the system at this time is:
Figure FDA0002301740160000062
c) in the current state of the system is
Figure FDA0002301740160000063
If the system takes the decision a 1, the next possible system state is j9＝<Nl,Nh+1,el>，j10＝<Nl,Nh+1,eh>，j11＝<Nl-1,Nh+1,ef>，NlNot less than 1 and j12＝<Nl,Nh,ef>(ii) a The state transition probability q (j | s, a) of the system at this time is:
Figure FDA0002301740160000064
and 8: calculating the maximum expected long-term discount yield:
first, the expected discount yield z (s, a) for the duration τ (s, a) between two decision points is calculated by applying the discount yield model of the markov decision process:
Figure FDA0002301740160000065
wherein the content of the first and second substances,
Figure FDA0002301740160000066
representing a desired discount value between two decision points; tau is1α represents a discount factor;
the maximum expected long-term discount yield of the system when the system state is s is obtained by the formula:
Figure FDA0002301740160000067
wherein upsilon (j) represents the maximum expected discount benefit of the system when the system state is j;
Figure FDA0002301740160000068
finding a finite constant ω such that it satisfies ω ═ λl+λh+K*max(μl,μh)<Infinity, simultaneously order
Figure FDA0002301740160000069
The maximum expected long-term discount yield after normalization is obtained as:
Figure FDA00023017401600000610
wherein
Figure FDA0002301740160000071
At the same time, the normalized state transition probability is,
Figure FDA0002301740160000072
and step 9: deriving the delay probability P of ALMRAMDelaying：
Figure FDA0002301740160000073
Simultaneously satisfies 0 ≤ αlNl+αhNhK or less, wherein
Figure FDA0002301740160000074
Figure FDA0002301740160000075
In system state separately for AMLRAS<Nl,Nh,el>And<Nl,Nh,eh>the corresponding decision to be taken when taking,
Figure FDA0002301740160000076
and
Figure FDA0002301740160000077
optimizing and distributing the steady-state probability of the management model for the anti-money laundering resource;
step 10: when delay probability PDelayingWhen the suspicious transaction report is smaller than a set threshold value, immediately analyzing the suspicious transaction report, and then allocating one or more anti-money laundering resource units to the system resource management functional module to analyze and investigate the suspicious transaction report; otherwise, the suspicious transaction report is delayed for analysis processing.
CN201911224355.8A 2019-12-04 2019-12-04 Financial anti-money laundering cloud computing resource optimal allocation system and method based on SMDP Pending CN111127201A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911224355.8A CN111127201A (en) 2019-12-04 2019-12-04 Financial anti-money laundering cloud computing resource optimal allocation system and method based on SMDP

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911224355.8A CN111127201A (en) 2019-12-04 2019-12-04 Financial anti-money laundering cloud computing resource optimal allocation system and method based on SMDP

Publications (1)

Publication Number Publication Date
CN111127201A true CN111127201A (en) 2020-05-08

Family

ID=70497352

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911224355.8A Pending CN111127201A (en) 2019-12-04 2019-12-04 Financial anti-money laundering cloud computing resource optimal allocation system and method based on SMDP

Country Status (1)

Country Link
CN (1) CN111127201A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112420189A (en) * 2020-10-22 2021-02-26 魏忠钰 Hierarchical disease diagnosis system, disease diagnosis method, device and apparatus
CN114490626A (en) * 2022-04-18 2022-05-13 成都数融科技有限公司 Financial information analysis method and system based on parallel computing

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109345375A (en) * 2018-09-27 2019-02-15 中国银行股份有限公司 A kind of suspicious money laundering Activity recognition method and device

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109345375A (en) * 2018-09-27 2019-02-15 中国银行股份有限公司 A kind of suspicious money laundering Activity recognition method and device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
XINTAO HONG ET AL: "Peer to Peer Anti-Money Laundering Resource Allocation Based on Semi-Markov Decision Process", 《2015 IEEE GLOBAL COMMUNICATIONS CONFERENCE (GLOBECOM)》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112420189A (en) * 2020-10-22 2021-02-26 魏忠钰 Hierarchical disease diagnosis system, disease diagnosis method, device and apparatus
CN114490626A (en) * 2022-04-18 2022-05-13 成都数融科技有限公司 Financial information analysis method and system based on parallel computing
CN114490626B (en) * 2022-04-18 2022-08-16 成都数融科技有限公司 Financial information analysis method and system based on parallel computing

Similar Documents

Publication Publication Date Title
Kallus et al. Residual unfairness in fair machine learning from prejudiced data
CN109461078A (en) A kind of abnormal transaction identification method and system based on funds transaction network
US20170262852A1 (en) Database monitoring system
CN109670936A (en) Loan examination &amp; approval processing method, platform, equipment and computer readable storage medium
CN110380897A (en) Network security situation awareness model and method based on improved BP
CN109191283A (en) Method for prewarning risk and system
CN109344583B (en) Threshold determination and body verification method and device, electronic equipment and storage medium
CN104915600B (en) A kind of Android application securitys methods of risk assessment and device
CN109377339B (en) Auxiliary screening method and system for suspicious transaction cases
CN109447783A (en) Credit method, apparatus, terminal device and storage medium
Buchak et al. Do mortgage lenders compete locally? Implications for credit access
CN111127201A (en) Financial anti-money laundering cloud computing resource optimal allocation system and method based on SMDP
Jiang et al. Deciphering big data in consumer credit evaluation
CN107257419B (en) One kind quantifying estimation method based on Bayesian analysis interpersonal relationships
US20150262184A1 (en) Two stage risk model building and evaluation
Hoi Financial development and income inequality in Vietnam: An empirical analysis
Ardielli Use of TOPSIS method for assessing of good governance in European Union countries
CN113450011A (en) Task allocation method and device
CA3110789A1 (en) Systems and methods for determining risk of identity fraud based on multiple fraud detection models
WO2022143431A1 (en) Method and apparatus for training anti-money laundering model
Akighir et al. An Empirical Analysis of the Impact of Agency Banking on Financial Inclusion in Benue State, Nigeria: Implications for Economic Activities
CN113408967B (en) Method and device for evaluating power one-network-through multi-state data sharing risk
CN114493864A (en) Capital big data based anomaly detection system and method
CN113822751A (en) Online loan risk prediction method
Chen et al. Digital Lending and Financial Well-Being: Evidence from a Developing Economy

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
CB03 Change of inventor or designer information
CB03 Change of inventor or designer information

Inventor after: Hong Xintao

Inventor after: Liang Hongbin

Inventor before: Liang Hongbin

Inventor before: Hong Xintao

RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200508