Machine learning-based import and export enterprise risk assessment method

Abstract

The invention discloses an import and export enterprise risk assessment method based on machine learning, and the method comprises the following steps: constructing an enterprise risk general situation database according to historical monitoring data, and carrying out the statistics of the total number of all items; generating a time weight and a risk weight of each type of items; calculating the risk value of each risk item for the overall item of the enterprise; dividing risk levels of the enterprises according to the corresponding risk values; calculating the membership degree of the enterprise risk in each risk level; and judging whether the enterprise is in a qualified range according to the membership matrix and feeding back. According to the invention, through historical monitoring data, an enterprise risk general situation database, a time risk double attention module, an enterprise risk assessment module, a risk grade division module, a first-level fuzzy comprehensive evaluation model and a feedback module, risk analysis of a target enterprise is realized; therefore, whether the enterprise has a risk can be judged according to the risk value of the target enterprise.

Description

Technical Field

[0001] The present invention relates to the field of computer and machine learning technology, and in particular to a risk assessment method and system for import and export enterprises based on machine learning. Background Art

[0002] With the development of globalization, the market exchanges between countries around the world are becoming closer and closer, and product exports have become an important means for countries to develop their economies. Today, China's export market continues to expand and has become one of the world's largest exporters. With the expansion of the international market, important cooperation between countries has become economic dependence on each other.

[0003] At present, product exports are an important source of national economy. With the continuous advancement of globalization, product exports will become more and more dependent on national economy. Export and import trade has become the most common economic interaction between countries. However, the laws, regulations, taxes and other contents involved are becoming more and more complicated. If the import and export enterprises are investigated by the public security organs and the industrial and commercial administrative departments for engaging in certain illegal activities, or there are high-risk situations such as capital rupture and bond default, then the above-mentioned enterprises with operating risks can be regarded as risky enterprises. Other enterprises that participate in the business activities of risky enterprises and have interests transferred with risky enterprises may also suffer from financial losses caused by the capital rupture of risky enterprises, and the risks of personnel suspected of crimes. It can be seen that the enterprise channels selected for import and export commodities are crucial, and determining the risk of import and export enterprises has become a major problem. For the above problems, the current technology has not yet proposed an effective solution.

[0004] Machine learning technology can simulate human thinking and use self-learning to achieve data analysis, decision-making, optimization and prediction. This learning ability is applied in risk assessment, especially in quantifying risks. Fuzzy comprehensive evaluation is a commonly used machine learning method. The fuzzy comprehensive evaluation model can handle complex multi-index problems. In actual decision-making problems, multiple indicators are often involved, and there may be certain correlations and mutual influences between these indicators. Traditional evaluation methods can only consider a few indicators and it is difficult to fully grasp the essence of the problem. The fuzzy comprehensive evaluation model can integrate the information of different indicators, comprehensively consider the weights and correlations of each indicator, and thus conduct more comprehensive decision analysis and evaluation.

[0005] Due to the large update interval of enterprise data, the data used in the evaluation model is not necessarily the current real-time data of the enterprise, and there is a certain ambiguity. For the ambiguity problem of enterprise risk assessment, it is very difficult to use traditional mathematical theory to accurately define the dynamic characteristics of customs risk assessment of import and export enterprises. This patent combines fuzzy mathematics to construct a risk assessment model, which can form a simpler and more efficient import and export enterprise risk assessment method that does not rely on internal enterprise data and customs historical behavior data. While reducing the difficulty of obtaining indicators, it ensures the universality of the assessment for all enterprises. Summary of the invention

[0006] In view of this, the purpose of the present invention is to provide an import and export enterprise risk assessment method based on machine learning. Risk analysis of the target enterprise is realized through historical monitoring data, enterprise risk profile database, time risk dual attention module, enterprise risk assessment module, risk level classification module, first-level fuzzy comprehensive evaluation model and feedback module, so that the risk of the target enterprise can be judged according to the risk value of the target enterprise.

[0007] The objective of the present invention is achieved through the following technical solutions:

[0008] The present invention provides a risk assessment method for import and export enterprises based on machine learning, comprising the following steps:

[0009] Step S1: Build an enterprise risk profile database based on historical monitoring data and count the total number of each item;

[0010] Step S2: Input the enterprise information to be evaluated in the enterprise risk profile database into the time-risk dual attention module to generate the time weight and risk weight of each type of matter;

[0011] Step S3: construct an enterprise risk value assessment module to calculate the risk value of each risk item for the overall enterprise items;

[0012] Step S4: construct a risk level classification module to classify the risk level of the enterprise according to the corresponding risk value;

[0013] Step S5: construct a first-level fuzzy comprehensive evaluation model to generate a membership matrix and calculate the membership of the enterprise risk in each risk level;

[0014] Step S6: Determine whether the enterprise is within the qualified range based on the membership matrix and provide feedback to relevant personnel.

[0015] Further, the step S1 includes the following sub-steps:

[0016] Step S101: pre-process the historical monitoring data and classify the various matters of the enterprise. The types of matters classified include but are not limited to the list of dishonest entities, abnormal operations, administrative penalties and credit commitments;

[0017] Step S102: Count the number of each type of items, and count the number of each item of each enterprise;

[0018] Step S103: storing the above four types of items of a single enterprise information into the same set of data, and storing all enterprise information into the enterprise risk profile database in the same way;

[0019] Further, the step S2 specifically includes the following sub-steps:

[0020] Step S201: construct a temporal risk dual attention module;

[0021] Step S202: Calculate the attenuation coefficient of the data input into the time risk dual attention module according to the historical information;

[0022] Step S203: Design a time weight algorithm according to the attenuation coefficient to calculate the time weight of each event; the expression of the time weight algorithm is:

[0023]

[0024] in, represents the time weight of the i-th event, T represents the meaning of time, k is the attenuation coefficient of the time weight, and Y i is the number of years that have passed since the occurrence of the i-th event;

[0025] Step S204: Design a risk weight algorithm to calculate the data input into the risk attention module to obtain the individual weight of each risk item.

[0026] Furthermore, the risk weight algorithm described in step S204 calculates the data input into the risk attention module to obtain the individual weights of each risk item. The weight algorithm expressions of the three risk items are respectively:

[0027]

[0028]

[0029]

[0030] Among them, W S represents the weight of the dishonest subject list, W J represents the weight of abnormal operation, W X represents the weight of administrative punishment, N S is the number of items on the list of all corporate dishonest entities, N J is the number of abnormal business events of all enterprises, N Xis the number of administrative penalties for all enterprises, R is the total number of risk items, r i is the number of risk items of category i;

[0031] Among them, AC () is the permutation and combination product addition operation, and the specific operation method is as follows:

[0032] AC(A, B, C)=A×B+B×C+A×C+A×B×C

[0033] Among them, A, B, and C are arbitrary real numbers.

[0034] Further, the step S3 specifically includes the following sub-steps:

[0035] Step S301: Assign the time weight of each event and the total number of credit commitments of the enterprises to be assessed N Cr Bonus units input into the enterprise risk assessment module;

[0036] Step S302: construct a positive event algorithm in the scoring unit to calculate the total bonus value of the enterprise; the expression of the positive event algorithm is:

[0037]

[0038] in, is the total bonus value of the enterprise, N Cr is the total number of credit commitments of the enterprise, Y i is the number of years that have passed since the occurrence of the i-th event;

[0039] Step S303: Assign the time weight of each event 、Number of negative items of the enterprise to be evaluated S 、N J 、N X And the weight of each negative item W S , W J , W X , input the deduction unit;

[0040] Step S304: constructing a negative event algorithm in the deduction unit to calculate the deduction value of the enterprise;

[0041] Furthermore, in step S304, in constructing a negative event algorithm in the deduction unit to calculate the deduction value of the enterprise, the expression of the negative event algorithm is:

[0042]

[0043]

[0044]

[0045] Among them, N S is the total number of the enterprise’s dishonest entities list, N J is the total number of abnormal operations of the enterprise, N X is the total number of administrative penalties imposed on the enterprise, is the time weight of the ith event, PTS S Points deducted for items on the list of dishonest subjects, PTS J Points deducted for abnormal business operations, PTS X Points deducted for administrative penalties.

[0046] Further, step S4 specifically includes:

[0047] The existing risky enterprises are divided into four levels, namely low risk, medium risk, medium-high risk and high risk;

[0048] The degree of risk is divided according to the risk value: enterprises with a risk value less than 80 are high-risk enterprises, enterprises with a risk value less than 90 but greater than or equal to 80 are medium-high risk enterprises, enterprises with a risk value less than 97 but greater than or equal to 90 are medium-risk enterprises, and enterprises with a risk value greater than or equal to 97 are low-risk enterprises.

[0049] The median risk value of each risk level is calculated according to the score threshold of each risk level, and the results are ranked from low risk to high risk in percentage as follows: the median risk value of low risk is (100+97) / 2, which is 98.5; the median risk value of medium risk is (97+90) / 2, which is 93.5; the median risk value of medium-high risk is (90+80) / 2, which is 85; and the median risk value of high risk is (80+0) / 2, which is 40.

[0050] Further, the step S5 specifically includes the following sub-steps:

[0051] Step S501: setting a comment set according to each risk level;

[0052] Step S502: Calculate the evaluation criteria of each comment set by integrating the bonus values ​​of the bonus units; the calculation formula is as follows:

[0053] std m =mid m +

[0054] Among them, std m is the evaluation criteria for the mth review set, mid m is the median risk value of the mth comment set, The total bonus points for the enterprise;

[0055] Step S503: designing a trapezoidal distribution membership function to calculate a fuzzy comprehensive evaluation matrix;

[0056] Step S504: combining the design weight matrix with the fuzzy comprehensive evaluation matrix to generate a membership matrix;

[0057] Step S505: Generate the membership degree of enterprise risk in each risk level from the membership matrix;

[0058] Furthermore, in step S503, a trapezoidal distribution membership function is designed to calculate the fuzzy comprehensive evaluation matrix, and the specific formula of the trapezoidal distribution membership function is as follows:

[0059]

[0060]

[0061]

[0062]

[0063] a m =100-std m

[0064] Among them, f ij is the initial membership of the i-th negative item in the j-th review set, PTS i is the deduction score of the ith negative item, i.e., the deduction score PTS of the dishonest subject list item calculated in step S304 S , PTS deduction points for abnormal business operations J And the deduction points for administrative penalties PTS S , a m is the membership value of the mth comment set, std m is the evaluation criteria for the mth review set;

[0065] Its fuzzy comprehensive evaluation matrix is:

[0066]

[0067] The step S504 designs a weight matrix and combines it with the fuzzy comprehensive evaluation matrix to generate a membership matrix, and the membership matrix calculation formula is as follows:

[0068] B=A×F

[0069] Among them, B is the membership matrix, which has one row and four columns, F is the fuzzy comprehensive evaluation matrix, which has three rows and four columns, and A is the weight matrix, which has one row and three columns. Its specific composition is as follows:

[0070]

[0071] The step S505 generates the degree of membership of the enterprise risk in each risk level from the degree of membership matrix, and the degree of membership matrix is ​​as follows:

[0072]

[0073] Among them, b1 is the degree of membership of the enterprise risk level as low risk, b2 is the degree of membership of the enterprise risk level as medium risk, b3 is the degree of membership of the enterprise risk level as medium-high risk, and b4 is the degree of membership of the enterprise risk level as high risk.

[0074] Further, step S6 determines whether the enterprise is within the qualified range according to the membership matrix and feeds back to relevant personnel, which specifically includes the following sub-steps:

[0075] Step S601: inputting the membership matrix of the enterprise to be evaluated into the early warning unit in the feedback module;

[0076] Step S602: The early warning unit determines whether to issue an early warning based on the membership matrix. If the enterprise risk level is high, the process proceeds to S603; otherwise, the process proceeds to S604.

[0077] Step S603: Feedback the information of the enterprise to relevant processing personnel and conduct a careful review of the enterprise;

[0078] Step S604: Display the risk affiliation of the enterprise and the total number of each item.

[0079] In step S602, the early warning unit determines whether to issue an early warning according to the risk level. If the high risk affiliation of the enterprise is greater than 50% (ie 0.5), the process proceeds to S603; otherwise, the process proceeds to S604.

[0080] The beneficial effects of the present invention include:

[0081] (1) The first-level fuzzy comprehensive evaluation model constructed by the present invention adopts a trapezoidal distribution membership function. It takes the customs data of import and export enterprises as the starting point, collects the enterprise information that is public and easier to obtain from import and export enterprises, and constructs a customs risk assessment model, thus forming a simpler and more efficient import and export enterprise risk assessment method that is independent of the internal data of the enterprise and the historical behavior data of customs. While reducing the difficulty of obtaining indicators, it ensures the universality of the assessment for all enterprises.

[0082] (2) The present invention assigns different weights to each negative event by designing a risk weight algorithm for each negative event, thereby increasing the differences between each negative event to weigh and calculate the risk level brought by each negative event to the enterprise, thereby improving the accuracy of enterprise risk assessment and providing detailed enterprise data for relevant departments;

[0083] (3) The present invention can accurately predict the risk level of the target enterprise by calculating the enterprise risk value through real data. The calculation method used is derived from the current judgment basis for the cooperation conditions of state-owned enterprises, and has an update and iteration mechanism to a certain extent, eliminating the errors caused by manual input, and the judgment basis is more scientific and accurate. In addition, it can reduce labor costs and increase the profits of enterprises;

[0084] (4) The present invention designs a time weight algorithm to evaluate the weights of events occurring in each year, wherein the attenuation coefficient k of the time weight is added, and its value can be dynamically adjusted according to the different institutional conditions each year to balance the impact of various events on the enterprise each year. The risk assessment of the enterprise has a certain degree of compensability and conducts risk assessment from a global perspective, which is suitable for most enterprises and improves the universality of the model.

[0085] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description and the preceding claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0087] Figure 1 It is a schematic diagram of the process of the present invention;

[0088] Figure 2 is a system structure diagram of the present invention;

[0089] Figure 3 This is the data storage format of the enterprise Q to be evaluated in an embodiment of the present invention.

[0090] Figure 4 This is a structural diagram of the time-risk dual attention module of the present invention;

[0091] Figure 5 This is a structural diagram of the enterprise risk value assessment module of the present invention. DETAILED DESCRIPTION

[0092] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments are only for illustrating the present invention, rather than for limiting the protection scope of the present invention.

[0093] The schematic diagram of the process of the present invention is as follows Figure 1 As shown, the system structure diagram of the present invention is as shown Figure 2As shown, this embodiment provides an import and export enterprise risk assessment method based on machine learning, which specifically includes the following steps:

[0094] Step S1: Build an enterprise risk profile database based on historical monitoring data and count the total number of each item;

[0095] In this embodiment, step S1 specifically includes the following steps:

[0096] Step S101: pre-process the historical monitoring data and classify the enterprise's matters into a list of dishonest entities, abnormal operations, administrative penalties, and credit commitments;

[0097] Step S102: Count the number of each type of items, and count the number of each item of each enterprise;

[0098] Step S103: storing the above four items of information of a single enterprise into the same set of data, and storing all enterprise information into the enterprise risk profile database in the same way;

[0099] Among them, step S102 counts the total number of each item and counts the independent number of each item of each enterprise. The number of each item shall be based on the official data of "Credit China".

[0100] In this embodiment, step S102 counts the total number of each item and counts the independent number of each item of each enterprise, and the number of each item is based on the official data of Credit China;

[0101] Among them, as of March 14, 2021, a total of 442,000 pieces of information on the list of dishonest subjects, 8.812 million pieces of information on the list of abnormal operations, and 1.047 million administrative penalties have been reported across the country;

[0102] In this embodiment, step S103 stores the above four items of single enterprise information in the same group of data, and stores all enterprise information in the enterprise risk profile database in the same way. The historical monitoring data stored in the enterprise risk profile database is stored in the form of a data table. Taking the enterprise Q information to be evaluated in this embodiment as an example, its storage format is as follows: Figure 3 As shown;

[0103] The horizontal axis of the table represents the name of the enterprise's event category, the vertical axis of the table represents the year in which the event occurred, and the intersection of the horizontal and vertical axes of the table represents the number of events of the enterprise to be evaluated.

[0104] S2: Input the enterprise information to be evaluated in the enterprise risk profile database into the time-risk dual attention module to generate the time weight and risk weight of each type of matter;

[0105] In this embodiment, step S2 specifically includes the following steps:

[0106] Step S201: construct a temporal risk dual attention module;

[0107] Step S202: Calculate the attenuation coefficient of the data input into the time attention module according to the historical information;

[0108] Step S203: designing a time weight algorithm according to the attenuation coefficient to calculate the time weight of each event;

[0109] Step S204: Design a risk weight algorithm to calculate the data input into the risk attention module to generate a separate weight for each risk item.

[0110] Among them, the time-risk dual attention module consists of a time attention unit and a risk attention unit, which respectively calculate the time weight and the weight of various risk items. Its structure diagram is as follows: Figure 4 As shown;

[0111] In this embodiment, step S202 calculates the attenuation coefficient of the data input into the time attention module according to the historical information, and the attenuation coefficient needs to be calculated according to the statute of limitations for prosecution of corporate violations of laws and regulations;

[0112] Among them, as of May 17, 2023, because the relevant laws of my country stipulate that the statute of limitations for corporate illegal prosecution is 20 years, the attenuation coefficient k can be calculated based on the severity of the average annual decrease by a certain coefficient: k=1 / 20, that is, the attenuation coefficient k in the time weight algorithm is 0.05;

[0113] In this embodiment, in step S203, a time weight algorithm is designed according to the attenuation coefficient to calculate the time weight of each event. The expression of the time weight algorithm is:

[0114]

[0115] in, represents the time weight of the i-th event, T represents the meaning of time, k is the attenuation coefficient of the time weight, and Y i is the number of years that have passed since the occurrence of the i-th event;

[0116] In this embodiment, substituting k=0.05 into the above formula, the expression of the time weight algorithm can be obtained as follows:

[0117]

[0118] In this embodiment, in step S204, a risk weight algorithm is designed to calculate the data input into the risk attention module to generate a separate weight for each risk item. The expressions of the weight algorithm for the three risk items are:

[0119]

[0120]

[0121]

[0122] Among them, W S represents the weight of the dishonest subject list, W J represents the weight of abnormal operation, W X represents the weight of administrative punishment, N S is the number of items on the list of all corporate dishonest entities, N J is the number of abnormal business events of all enterprises, N X is the number of administrative penalties for all enterprises, R is the total number of risk items, r i is the number of risk items of category i;

[0123] Among them, AC () is the product addition operation of permutation and combination, and the specific operation method is as follows:

[0124] AC(A, B, C)=A×B+B×C+A×C+A×B×C

[0125] Among them, A, B, and C are arbitrary real numbers.

[0126] In this embodiment, the number of items on the list of all corporate dishonest entities is N S The number of abnormal business events of all enterprises is N J The number of administrative penalties for all enterprises is 8.812 million. X is 1.047 million, then the weighting algorithm expressions of the three risk items that can be obtained by substituting the above three data are:

[0127]

[0128]

[0129]

[0130] The weight W of the severity of the items on the list of dishonest subjects is obtained. S =0.679, weight of severity of abnormal business events W J =0.034, the weight of the severity of administrative penalties W X =0.287.

[0131] Step S3: construct an enterprise risk value assessment module to calculate the risk value of each risk item for the overall enterprise items;

[0132] In this embodiment, step S3 specifically includes the following steps:

[0133] Step S301: Assign the time weight of each event and the total number of credit commitments of the enterprises to be assessed N Cr Bonus units input into the enterprise risk assessment module;

[0134] Step S302: constructing a positive event algorithm in the scoring unit to calculate the total scoring value of the enterprise;

[0135] Step S303: Assign the time weight of each event 、Number of negative items of the enterprise to be evaluated S 、N J 、N X And the weight of each negative item W S , W J , W X , input the deduction unit;

[0136] Step S304: constructing a negative event algorithm in the deduction unit to calculate the deduction value of the enterprise;

[0137] Among them, the enterprise risk value assessment module of the import and export enterprise risk assessment method and system based on machine learning of the present invention is shown in FIG. Figure 5 As shown, the module consists of a score adding unit and a score deduction unit, which respectively calculate the total score adding value and the total score deduction value;

[0138] In this embodiment, in step S302, a positive event algorithm is constructed in the scoring unit to calculate the total score of the enterprise. The expression of the positive event algorithm is:

[0139]

[0140] in, is the total bonus value of the enterprise, N Cr is the total number of credit commitments of the enterprise, Y i is the number of years that have passed since the occurrence of the i-th event;

[0141] In this embodiment, if Figure 3 As shown, the total number of corporate credit commitments N Cr is 2, the years are 2022 and 2019, and the years from now are 1 year and 4 years respectively. Substituting the above data, the expression of the positive event algorithm is:

[0142] That is, the total bonus value of the enterprise can be obtained by calculation is 1.75;

[0143] In this embodiment, in step S304, a negative event algorithm is constructed in the deduction unit to calculate the deduction value of the enterprise. The expression of the negative event algorithm is:

[0144]

[0145]

[0146]

[0147] Among them, N S is the total number of the enterprise’s dishonest entities list, N J is the total number of abnormal operations of the enterprise, N X is the total number of administrative penalties imposed on the enterprise, is the time weight of the ith event, PTS S Points deducted for items on the list of dishonest subjects, PTS J Points deducted for abnormal business operations, PTS X Points deducted for administrative penalties.

[0148] In this embodiment, if Figure 3 As shown, the total number of the defaulting entities on the list of enterprise Q is N S is 4, the years are 2022, 2022, 2020 and 2019, and the years from now are 1 year, 1 year, 3 years and 4 years respectively; the total number of abnormal business operations N J is 4, the years are 2023, 2021, 2018 and 2018 respectively, and the years from now are 0, 2, 5 and 5 respectively; the total number of administrative penalties for enterprise Q is N X is 3, the years are 2020, 2020 and 2018 respectively, and the years since then are 3 years, 3 years and 5 years respectively. Substituting the above data, the deduction points for each negative matter are:

[0149]

[0150]

[0151]

[0152] PTS points deduction for items on the list of dishonest subjects S The PTS is 3.55, and the deduction score for abnormal business matters J The PTS is 3.4, the administrative penalty deduction score X is 2.45.

[0153] Step S4: construct a risk level classification module to classify the risk level of the enterprise according to the corresponding risk value;

[0154] In this embodiment, step S4 specifically includes the following steps:

[0155] Step S401: query the latest "White Paper on Risk Index of Chinese Listed Companies" through the risk level classification module;

[0156] Step S402: Obtain the boundary risk value for dividing risk levels in the latest system from the "White Paper on Risk Index of Chinese Listed Companies";

[0157] Step S403: Calculate the median risk value of each risk level according to the score threshold of each risk level;

[0158] In this embodiment, step S401 is to query the latest "White Paper on Risk Index of Chinese Listed Companies" by the risk level classification module, wherein the "White Paper on Risk Index of Chinese Listed Companies" has the most authoritative evaluation indicators for enterprise risk assessment;

[0159] In this embodiment, step S402 obtains the boundary risk value for dividing the risk level from the "White Paper on Risk Index of Chinese Listed Companies", which divides existing risky enterprises into four levels, namely low risk, medium risk, medium-high risk and high risk;

[0160] Among them, as of now, the "White Paper on Risk Index of Chinese Listed Companies" divides the degree of risk according to the risk value: a risk value of less than 80 points is a high-risk enterprise, a risk value of less than 90 points but greater than or equal to 80 points is a medium-high risk enterprise, a risk value of less than 97 points but greater than or equal to 90 points is a medium-risk enterprise, and a risk value greater than or equal to 97 points is a low-risk enterprise;

[0161] In this embodiment, step S403 calculates the median risk value of each risk level according to the score threshold of each risk level, calculates it in percentage and arranges the results from low risk to high risk as follows: the median risk value of low risk is (100+97) / 2, which is 98.5, the median risk value of medium risk is (97+90) / 2, which is 93.5, the median risk value of medium-high risk is (90+80) / 2, which is 85, and the median risk value of high risk is (80+0) / 2, which is 40.

[0162] Step S5: construct a first-level fuzzy comprehensive evaluation model to generate a membership matrix and calculate the membership of the enterprise risk in each risk level;

[0163] In this embodiment, step S5 specifically includes the following steps:

[0164] Step S501: setting a comment set according to each risk level;

[0165] Step S502: Calculate the evaluation criteria of each comment set by integrating the added score of the added score unit;

[0166] Step S503: designing a trapezoidal distribution membership function to calculate a fuzzy comprehensive evaluation matrix;

[0167] Step S504: combining the design weight matrix with the fuzzy comprehensive evaluation matrix to generate a membership matrix;

[0168] Step S505: Generate the membership degree of enterprise risk in each risk level from the membership matrix;

[0169] In this embodiment, step S501 sets a comment set according to each risk level. In the enterprise risk assessment, the comment set m is 4, which are low risk, medium risk, medium-high risk and high risk;

[0170] In this embodiment, step S502 integrates the score value of the score adding unit to calculate the evaluation criteria of each comment set, and the calculation formula is as follows:

[0171] std m =mid m +

[0172] Among them, std m is the evaluation criteria for the mth review set, mid m is the median risk value of the mth comment set, The total bonus value of the enterprise;

[0173] In this embodiment, the median risk values ​​of the comment sets are 98.5, 93.5, 85, and 40, respectively. By calculating the above formula, the evaluation standard std of each comment set is 100.25, 95.25, 86.75, and 41.75, respectively;

[0174] Step S503 designs a trapezoidal distribution membership function to calculate the fuzzy comprehensive evaluation matrix. The specific formula of the trapezoidal distribution membership function is as follows:

[0175]

[0176]

[0177]

[0178]

[0179] a m =100-std m

[0180] Among them, f ij is the initial membership of the i-th negative item in the j-th review set, PTS iis the deduction score of the ith negative item, i.e., the deduction score PTS of the dishonest subject list item calculated in step S304 S , PTS deduction points for abnormal business operations J And the deduction points for administrative penalties PTS X , a m is the membership value of the mth comment set, std m is the evaluation criteria for the mth review set;

[0181] In this embodiment, the membership values ​​a1, a2, a3, and a4 of the comment set can be calculated to be -0.25, 4.75, 13.25, and 58.25, respectively. Then the specific formula of the trapezoidal distribution membership function can be updated as follows:

[0182]

[0183]

[0184]

[0185]

[0186] In this embodiment, step S503 designs a trapezoidal distribution membership function to calculate a fuzzy comprehensive evaluation matrix, and the fuzzy comprehensive evaluation matrix is:

[0187]

[0188] In this embodiment, the fuzzy comprehensive evaluation matrix has three rows and four columns, and the specific format is as follows:

[0189]

[0190] In this embodiment, the fuzzy comprehensive evaluation matrix can be obtained by calculating the above formula:

[0191]

[0192] Step S504 designs a weight matrix and combines it with the fuzzy comprehensive evaluation matrix to generate a membership matrix. The membership matrix calculation formula is as follows:

[0193] B=A×F

[0194] Among them, B is the membership matrix, which has one row and four columns, F is the fuzzy comprehensive evaluation matrix, which has three rows and four columns, and A is the weight matrix, which has one row and three columns. Its specific composition is as follows:

[0195]

[0196] In this embodiment, the weight matrix A is as follows:

[0197]

[0198] Step S505 generates the degree of membership of enterprise risk in each risk level from the degree of membership matrix, and the degree of membership matrix is ​​as follows:

[0199]

[0200] Among them, b1 is the degree of membership when the enterprise risk level is low risk, b2 is the degree of membership when the enterprise risk level is medium risk, b3 is the degree of membership when the enterprise risk level is medium-high risk, and b4 is the degree of membership when the enterprise risk level is high risk;

[0201] In this embodiment, the weight matrix B obtained by the above calculation is as follows:

[0202]

[0203] Among them, the enterprise's risk level is low risk and the degree of membership is 0.24676, and the enterprise's risk level is medium risk and the degree of membership is 0.75324.

[0204] S6: Determine whether the enterprise is within the qualified range based on the membership matrix and provide feedback to relevant personnel;

[0205] In this embodiment, step S6 specifically includes the following steps:

[0206] S601: Input the membership matrix of the enterprise to be evaluated into the early warning unit in the feedback module;

[0207] S602: The early warning unit determines whether to issue an early warning based on the membership matrix. If the enterprise risk level is high, it proceeds to S603; otherwise, it proceeds to S604;

[0208] S603: Feedback the information of the enterprise to the relevant processing personnel and conduct a careful review of the enterprise;

[0209] S604: Display the risk affiliation of the enterprise and the total number of each item.

[0210] In this embodiment, in step S602, the early warning unit determines whether to issue an early warning according to the risk level. If the high-risk affiliation of the enterprise is greater than 50% (i.e. 0.5), it proceeds to S603. Otherwise, it proceeds to S604. The customs shall strictly inspect the import and export products of high-risk enterprises, so it is necessary to issue an early warning for the assessed high-risk enterprise information.

[0211] Among them, the high risk membership of enterprise Q is 0, which is less than 50%, so it will not enter S603;

[0212] In this embodiment, in step S604, the display unit needs to display the total details of the risk level of the enterprise and the total details of the four events of the list of dishonest entities, abnormal operation, administrative penalties, and credit commitments of the enterprise;

[0213] Among them, the output of the display unit is: the risk level of enterprise Q is low risk, the membership degree is 0.24676, the risk level is medium risk, the membership degree is 0.75324, the total number of untrustworthy entities on the list is 4, which occurred 2 times in 2022, 1 time in 2020, and 1 time in 2019 respectively; the total number of enterprise operation abnormalities is 4, which occurred 1 time in 2023, 1 time in 2021, and 2 times in 2018 respectively; the total number of administrative penalties is 3, which occurred 2 times in 2020 and 1 time in 2018 respectively; the total number of enterprise credit commitments is 2, which occurred 1 time in 2022 and 1 time in 2019 respectively.

[0214] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0215] A person skilled in the art may understand that all or part of the steps of the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0216] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0217] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0218] Furthermore, the operations of the processes described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) that is executed collectively on one or more processors, by hardware, or a combination thereof. The computer program includes a plurality of instructions that may be executed by one or more processors.

[0219] Further, the method can be implemented in any type of computing platform that is operably connected to a suitable computer, including but not limited to a personal computer, a minicomputer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, an optical read and / or write storage medium, a RAM, a ROM, etc., so that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the process described herein. In addition, the machine-readable code, or portions thereof, can be transmitted via a wired or wireless network. When such media includes instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor, the invention described herein includes these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques of the present invention, the present invention also includes the computer itself.

[0220] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A risk assessment method for import and export enterprises based on machine learning, characterized by: The method comprises the following steps: Step S1: Build an enterprise risk profile database based on historical monitoring data and count the total number of each item; Step S2: Input the enterprise information to be evaluated in the enterprise risk profile database into the time-risk dual attention module to generate the time weight and risk weight of each type of matter; Step S3: construct an enterprise risk value assessment module to calculate the risk value of each risk item for the overall enterprise items; Step S4: construct a risk level classification module to classify the risk level of the enterprise according to the corresponding risk value; Step S5: construct a first-level fuzzy comprehensive evaluation model to generate a membership matrix, and calculate the membership of the enterprise risk in each risk level; Step S6: Determine whether the enterprise is within the qualified range based on the membership matrix and provide feedback to relevant personnel.

2. The method for risk assessment of import and export enterprises based on machine learning according to claim 1 is characterized by: The step S1 comprises the following sub-steps: Step S101: pre-processing the historical monitoring data, and classifying the various matters of the enterprise, the types of which include but are not limited to the list of dishonest entities, abnormal operations, administrative penalties, and credit commitments; Step S102: Count the number of each type of items, and count the number of each item of each enterprise; Step S103: The above four types of item quantities of a single enterprise information are stored in the same set of data, and all enterprise information is stored in the enterprise risk profile database in the same way.

3. The method for risk assessment of import and export enterprises based on machine learning according to claim 1 is characterized by: The step S2 specifically includes the following sub-steps: Step S201: constructing a time-risk dual attention module, which includes a time attention module and a risk attention module; Step S202: Calculate the attenuation coefficient of the data input into the time attention module according to the historical information; Step S203: Design a time weight algorithm according to the attenuation coefficient to calculate the time weight of each event; the expression of the time weight algorithm is: in, represents the time weight of the i-th event, T represents the meaning of time, k is the attenuation coefficient of the time weight, and Y i is the number of years that have passed since the occurrence of the i-th event; Step S204: Design a risk weight algorithm to calculate the data input into the risk attention module to obtain the individual weight of each risk item.

4. The method for risk assessment of import and export enterprises based on machine learning according to claim 3 is characterized by: The designed risk weight algorithm described in step S204 calculates the data input into the risk attention module to obtain the individual weights of each risk item. The expressions of the weight algorithms of the three risk items are respectively: Among them, W S represents the weight of the dishonest subject list, W J represents the weight of abnormal operation, W X represents the weight of administrative punishment, N S is the number of items on the list of all corporate dishonest entities, N J is the number of abnormal business events of all enterprises, N X is the number of administrative penalties for all enterprises, R is the total number of risk items, r i is the number of risk items of category i; Among them, AC( ) is the product addition operation of permutation and combination, and the specific operation method is as follows: AC(A, B, C)=A×B+B×C+A×C+A×B×C Among them, A, B, and C are arbitrary real numbers.

5. The method for risk assessment of import and export enterprises based on machine learning according to claim 1 is characterized in that: The step S3 specifically includes the following sub-steps: Step S301: Assign the time weight of each event and the total number of credit commitments of the enterprises to be assessed N Cr Bonus units input into the enterprise risk assessment module; Step S302: construct a positive event algorithm in the scoring unit to calculate the total score of the enterprise; the expression of the positive event algorithm is: in, is the total bonus value of the enterprise, N Cr is the total number of credit commitments of the enterprise, Y i is the number of years that have passed since the occurrence of the i-th event; Step S303: Assign the time weight of each event 、Number of negative items of the enterprise to be evaluated S 、N J 、N X And the weight of each negative item W S , W J , W X , input the deduction unit; Step S304: construct a negative event algorithm in the deduction unit to calculate the deduction value of the enterprise.

6. The method for risk assessment of import and export enterprises based on machine learning according to claim 5 is characterized by: In step S304, in constructing a negative event algorithm in the deduction unit to calculate the deduction value of the enterprise, the expression of the negative event algorithm is: Among them, N S is the total number of the enterprise’s dishonest entities list, N J is the total number of abnormal operations of the enterprise, N X is the total number of administrative penalties imposed on the enterprise, is the time weight of the ith event, PTS S Points deducted for items on the list of dishonest subjects, PTS J Points deducted for abnormal business operations, PTS X Points deducted for administrative penalties.

7. The method for risk assessment of import and export enterprises based on machine learning according to claim 1 is characterized by: Step S4 specifically includes: The existing risky enterprises are divided into four levels, namely low risk, medium risk, medium-high risk and high risk; The degree of risk is divided according to the risk value: a risk value less than 80 points is a high-risk enterprise, a risk value less than 90 points but greater than or equal to 80 points is a medium-high risk enterprise, a risk value less than 97 points but greater than or equal to 90 points is a medium-risk enterprise, and a risk value greater than or equal to 97 points is a low-risk enterprise; The median risk value of each risk level is calculated according to the score threshold of each risk level, and the results are ranked from low risk to high risk in percentage as follows: the median risk value of low risk is (100+97) / 2, which is 98.5; the median risk value of medium risk is (97+90) / 2, which is 93.5; the median risk value of medium-high risk is (90+80) / 2, which is 85; and the median risk value of high risk is (80+0) / 2, which is 40.

8. The method for risk assessment of import and export enterprises based on machine learning according to claim 1 is characterized by: The step S5 specifically includes the following sub-steps: Step S501: setting a comment set according to each risk level; Step S502: Calculate the evaluation criteria of each comment set by integrating the bonus values ​​of the bonus units; the calculation formula is as follows: standard m =mid m + Among them, std m is the evaluation criteria for the mth review set, mid m is the median risk value of the mth comment set, The total bonus points for the enterprise; Step S503: designing a trapezoidal distribution membership function to calculate a fuzzy comprehensive evaluation matrix; Step S504: combining the design weight matrix with the fuzzy comprehensive evaluation matrix to generate a membership matrix; Step S505: Generate the membership degree of enterprise risk in each risk level from the membership matrix.

9. The method for risk assessment of import and export enterprises based on machine learning according to claim 8 is characterized by: The step S503 designs a trapezoidal distribution membership function to calculate the fuzzy comprehensive evaluation matrix, and the specific formula of the trapezoidal distribution membership function is as follows: a m =100- std m Among them, f ij is the initial membership of the i-th negative item in the j-th review set, PTS i is the deduction score of the ith negative item, i.e., the deduction score PTS of the dishonest subject list item calculated in step S304 S , PTS deduction points for abnormal business operations J And the deduction points for administrative penalties PTS S , a m is the membership value of the mth comment set, std m is the evaluation criteria for the mth review set; Its fuzzy comprehensive evaluation matrix is: The step S504 designs a weight matrix and combines it with the fuzzy comprehensive evaluation matrix to generate a membership matrix, and the membership matrix calculation formula is as follows: B=A×F Among them, B is the membership matrix, which has one row and four columns, F is the fuzzy comprehensive evaluation matrix, which has three rows and four columns, and A is the weight matrix, which has one row and three columns. Its specific composition is as follows: The step S505 generates the degree of membership of the enterprise risk in each risk level from the degree of membership matrix, and the degree of membership matrix is ​​as follows: Among them, b1 is the degree of membership of the enterprise risk level as low risk, b2 is the degree of membership of the enterprise risk level as medium risk, b3 is the degree of membership of the enterprise risk level as medium-high risk, and b4 is the degree of membership of the enterprise risk level as high risk.

10. The method for risk assessment of import and export enterprises based on machine learning according to claim 1, characterized in that: Step S6 determines whether the enterprise is within the qualified range according to the membership matrix and gives feedback to relevant personnel, which specifically includes the following sub-steps: Step S601: inputting the membership matrix of the enterprise to be evaluated into the early warning unit in the feedback module; Step S602: The early warning unit determines whether to issue an early warning based on the membership matrix. If the enterprise risk level is high, the process proceeds to S603; otherwise, the process proceeds to S604. Step S603: Feedback the information of the enterprise to relevant processing personnel and conduct a careful review of the enterprise; Step S604: Display the risk affiliation of each enterprise and the total number of each item; In step S602, the early warning unit determines whether to issue an early warning according to the risk level. If the high risk affiliation of the enterprise is greater than 50% (ie 0.5), the process proceeds to S603; otherwise, the process proceeds to S604.

Images