Intelligent internal control test verification method and system

Abstract

The present invention discloses an intelligent internal control test verification method and system, which relates to the technical field of enterprise internal control verification, and includes the following steps: in the business scenario of a multi-level approval process, the actual detection information of all approval nodes is comprehensively collected to establish a complete analysis set; after each approval detection of the approval node is completed, the approval nodes that are not executed as expected in the approval process are preliminarily identified. The present invention realizes differentiated internal control by classifying nodes into high-risk approval nodes and normal approval nodes, inserts new approval links into high-risk nodes, and dynamically adjusts their frequency according to the risk level, improves security in high-risk scenarios, and avoids excessive intervention in low-risk scenarios. Through real-time monitoring and dynamic adjustment, logical conflicts and process loopholes are reduced, fund leakage and unauthorized operations are prevented, internal control resource optimization is realized, approval efficiency is improved, compliance and financial security of the enterprise are guaranteed, and strong support is provided for the stable operation of the enterprise.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise internal control verification, and in particular to an intelligent internal control test verification method and system. Background Art

[0002] The intelligent internal control test and verification system integrates automation technology and data analysis capabilities, aiming to verify the effectiveness and compliance of an enterprise's internal control. By collecting information such as key business processes, financial data, and operation logs, the system uses preset rules and machine learning models to identify potential risks and control deficiencies. The system has the function of dynamically generating test cases, simulating a variety of business scenarios, and real-time monitoring and feedback verification to achieve full coverage testing of internal control links. The system supports seamless docking with information systems such as ERP, reduces human intervention, and improves testing efficiency, thereby ensuring the robustness and transparency of the enterprise's internal control system.

[0003] In actual use, the system will automatically generate accurate test cases based on business processes and internal control standards. By analyzing the company's business logic, financial processes, authority management and control nodes, and combining internal control frameworks (such as COSO, SOX) or custom rules, the system dynamically creates test plans. The generated test cases cover a variety of scenarios such as procurement approval, fund payment, inventory management, etc., simulating potential business operations and abnormal situations. The system not only supports routine testing, but also generates special tests for new business changes and complex processes to ensure comprehensive coverage of all key links. After the test results are fed back in real time, the system can automatically optimize the rules and internal control strategies to continuously improve the accuracy and efficiency of the test.

[0004] The prior art has the following deficiencies:

[0005] In the existing intelligent internal control test verification system, the automatically generated test cases rely on preset rules and internal control logic. However, when the system matches rules in a complex approval process, logical loops may occur. In business scenarios with multi-level approval processes, if the duplicate detection logic between approval nodes bypasses each other, some key control links may not be verified as expected. This kind of logical conflict or coverage blind spot, especially in key processes such as fund approval and large payments, will cause some processes to be ignored. Once the key links are bypassed, unauthorized large payments or duplicate payments may be executed, leading to serious risks such as fund leakage.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0007] The purpose of the present invention is to provide an intelligent internal control test verification method and system, which establishes an analysis set, comprehensively collects and compares actual and expected detection information, identifies approval nodes that are not executed as expected, and extracts features from abnormal nodes, then inputs the machine learning model to achieve dynamic risk prediction and evaluation, and divides the nodes into high-risk and normal nodes to achieve differentiated internal control. For high-risk nodes, the system automatically inserts the approval link, and dynamically adjusts the insertion frequency according to the risk level, while ensuring the safety of high-risk scenarios, avoiding excessive intervention in low-risk situations, and improving approval efficiency. Through risk classification and internal control strengthening, risks such as fund leakage and repeated payment are effectively prevented, and internal control resources are optimized. Through real-time monitoring and adjustment, process transparency and compliance are guaranteed, ensuring that enterprises maintain efficient operations while improving security, and responding to risk changes in a timely manner, achieving continuous optimization and improvement of the internal control system, and supporting the long-term and stable development of enterprises, so as to solve the problems in the above-mentioned background technology.

[0008] In order to achieve the above object, the present invention provides the following technical solution: an intelligent internal control test verification method, comprising the following steps:

[0009] In the business scenario of multi-level approval process, the actual inspection information of all approval nodes is comprehensively collected to establish a complete analysis set;

[0010] After each approval check of the approval node is completed, the approval nodes that are not executed as expected in the approval process are preliminarily identified, and the approval nodes that are preliminarily identified as not executed as expected are marked and filtered out;

[0011] For the approval nodes that are initially identified as not being executed as expected, the actual test information after the approval test is completed is further compared and analyzed with the expected test information, key features are extracted from the compared data, and the extracted features are analyzed and processed;

[0012] The analyzed and processed feature data is input into the pre-trained machine learning model, and the actual detection situation of the approval node is predicted through the machine learning model to assess the potential risk of repeated detection logic bypassing each approval node.

[0013] According to the prediction results of the machine learning model, the approval nodes are classified into high-risk approval nodes and normal approval nodes;

[0014] For approval nodes classified as high-risk, take enhanced internal control measures and insert new approval links in high-risk approval nodes;

[0015] In order to ensure a balance between security and efficiency in the approval process, the insertion frequency of new approval links is dynamically adjusted according to the risk level of bypassing duplicate detection logic of approval nodes.

[0016] Preferably, after each approval test is completed, the specific steps of preliminarily identifying the approval nodes that are not executed as expected in the approval process, and marking and screening out the approval nodes that are preliminarily identified as not executed as expected are as follows:

[0017] After the approval process is completed, the actual detection information of each approval node in the analysis set is compared with the expected detection information one by one, and the approval operations that do not meet expectations are found through the rule engine;

[0018] After the comparison is completed, the approval nodes that do not meet expectations are screened out;

[0019] Once the approval nodes with abnormalities are initially identified, they will be marked and screened out to enter further analysis and processing flow.

[0020] Preferably, after each approval detection of the approval node is completed, key features are extracted from the compared data, wherein the extracted key features include the frequency of the approval node not being triggered when it is expected to be executed and the degree of deviation between the actual processing time of the approval node and the expected processing time. After acquisition, the frequency of the approval node not being triggered when it is expected to be executed and the degree of deviation between the actual processing time of the approval node and the expected processing time are analyzed and processed to generate an approval skip factor and an approval time anomaly factor respectively. The degree of lack of integrity in the approval link is measured by the approval skip factor, and the execution efficiency of the approval operation is evaluated by the approval time anomaly factor.

[0021] Preferably, after obtaining the approval skip factor and the approval time anomaly factor generated by analyzing and processing the extracted key features, the approval skip factor and the approval time anomaly factor are input into a pre-trained machine learning model to generate a detection logic bypass coefficient, and the detection logic bypass coefficient is used to evaluate the potential risk of repeated detection logic bypassing each approval node.

[0022] Preferably, the detection logic bypass coefficient generated by predicting the actual detection situation of the approval node through the pre-trained machine learning model is compared with the pre-set detection logic bypass coefficient reference threshold to classify the approval nodes. The specific classification steps are as follows:

[0023] If the detection logic bypass coefficient is greater than the detection logic bypass coefficient reference threshold, the approval node is classified as a high-risk approval node;

[0024] If the detection logic bypass coefficient is less than or equal to the detection logic bypass coefficient reference threshold, the approval node is classified as a normal approval node, and the normal approval node is a node where the process execution is normal.

[0025] Preferably, the specific steps of generating the approval skip factor after analyzing the frequency of the approval node not being triggered when it is expected to be executed are as follows:

[0026] Collect the actual detection information and expected detection information of the approval node, and calculate the time priority coefficient of the approval node. The calculation expression is:

[0027]

[0028] , where P time (t) is the time priority coefficient of the approval node, indicating the importance of the approval node in the process, t is the execution time of the current approval node, t0 is the key time point of the approval node, and α is the parameter for adjusting the priority change rate;

[0029] By analyzing the frequency and importance of approval nodes being skipped, the skip weight is calculated. The calculation expression of the skip weight is: W miss =R missed ·P time (t), where W miss is the skip weight, which comprehensively considers the skip frequency and time priority of the approval node. missed The skip frequency of the approval node is calculated as follows:

[0030]

[0031] , where N total N is the total number of times the approval node is expected to be triggered. missed The number of times the approval node was not actually triggered, that is, the number of times the approval node should be executed but was not executed;

[0032] The approval skip factor is calculated by skipping weight to quantify the degree of completeness loss in the approval process. The calculation expression is:

[0033]

[0034] , where I skip is the approval skip factor, and γ is the correction parameter, which is used to control the degree of correction of the approval skip factor by the non-skipped approval node.

[0035] Preferably, the specific steps of generating the approval time abnormality factor after analyzing the deviation between the actual processing time of the approval node and the expected processing time are as follows:

[0036] Define the time deviation function to quantify the difference between the actual processing time and the expected processing time of the approval node. The time deviation function serves as the basis for subsequent calculations. The function expression is: ΔT(t) = |T actual (t)-T expected(t)|, where ΔT(t) is the time deviation function, which represents the time error of the approval node at time t, T actual (t) is the actual processing time function of the approval node at time t, T expected (t) is the expected processing time function at the approval node at time t;

[0037] In order to evaluate the overall time deviation of the approval node, the time deviation function is integrated during the operation of the approval node to calculate the total accumulated time deviation. The calculation expression is:

[0038]

[0039] , where S(t) is the total time deviation accumulation, which means the total time deviation of the approval node in the entire operation time period, t 开始 is the start time of approval node processing, t 结束 The end processing time of the approval node;

[0040] After calculating the cumulative time deviation, the processing complexity of the approval node is further introduced to calculate the approval time anomaly factor to more comprehensively evaluate the efficiency of the approval operation. The calculation expression of the approval time anomaly factor is:

[0041]

[0042] , where Time Anomaly is the approval time abnormality factor, which is used to evaluate the abnormal degree of the execution efficiency of the approval node. C(t) is the processing complexity coefficient, which indicates the complexity of the approval node during the operation.

[0043]

[0044] , where D(τ) is the amount of dynamic processing data of the approval task at time τ, W(τ) is the task complexity weight function, which indicates the complexity of the approval matter, P is the authority level of the approver, which is used as the basis for adjusting the complexity, τ indicates a certain instantaneous time point in the integration process, and T expected The total expected processing time of the approval node.

[0045] Preferably, according to the risk level of the duplicate detection logic of the approval nodes bypassing each other, the insertion frequency of the new approval link is dynamically adjusted. The specific steps are as follows:

[0046] Calculate the risk exceedance ratio of high-risk approval nodes. The calculation expression is:

[0047]

[0048] , where R exceedis the risk exceedance ratio, which is used to measure the degree to which the detection logic bypass coefficient of the approval node exceeds the detection logic bypass coefficient reference threshold. Logic is the detection logic bypass coefficient, which indicates the actual risk level of the approval node. threshold Bypass coefficient reference threshold for detection logic;

[0049] After determining the risk exceedance ratio, the insertion frequency of the new approval link is dynamically adjusted according to the risk exceedance ratio. During regulation, a frequency adjustment factor is introduced to determine the insertion frequency of the new approval link. The calculation expression of the frequency adjustment factor is: F adjust =ψ·ln(1+R exceed ), where F adjust is the frequency adjustment factor, which represents the insertion frequency adjustment amount based on the actual risk level of the approval node, and ψ is the adjustment coefficient;

[0050] The new approval link insertion frequency is dynamically calculated based on the original approval link insertion frequency and the frequency adjustment factor. The calculation expression is: insert =f0·(1+F adjust ), where f insert is the new insertion frequency of the final application, indicating the actual insertion frequency of the new approval link under the current risk level, and f0 is the initial insertion frequency, indicating the insertion frequency of the approval link under the default condition.

[0051] An intelligent internal control test verification system, comprising a data collection module, an abnormal node identification module, a feature extraction and analysis module, a risk prediction module, a risk classification module, an internal control strengthening module and a dynamic control module;

[0052] The data collection module comprehensively collects the actual inspection information of all approval nodes in the business scenario of multi-level approval process and establishes a complete analysis set;

[0053] Abnormal node identification module: After each approval node inspection is completed, it preliminarily identifies the approval nodes that are not executed as expected in the approval process, and marks and screens out the approval nodes that are preliminarily identified as not executed as expected;

[0054] The feature extraction and analysis module further compares and analyzes the actual detection information after the approval detection is completed with the expected detection information for the approval nodes that are initially identified as not executed as expected, extracts key features from the compared data, and analyzes and processes the extracted features;

[0055] The risk prediction module inputs the analyzed and processed feature data into a pre-trained machine learning model, uses the machine learning model to predict the risk of the actual detection situation of the approval node, and evaluates the potential risk of repeated detection logic bypassing each approval node;

[0056] The risk classification module classifies approval nodes according to the prediction results of the machine learning model, dividing the approval nodes into high-risk approval nodes and normal approval nodes;

[0057] The internal control strengthening module takes strengthened internal control measures for approval nodes classified as high-risk and inserts new approval links into high-risk approval nodes;

[0058] The dynamic control module, in order to ensure the balance between security and efficiency in the approval process, dynamically adjusts the insertion frequency of new approval links according to the risk level of bypassing each other in the duplicate detection logic of approval nodes.

[0059] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0060] The present invention establishes a complete analysis set, comprehensively collects and compares actual detection information with expected detection information, and preliminarily identifies approval nodes that are not executed as expected. On this basis, key features are extracted from abnormal nodes, and their data is input into a pre-trained machine learning model to achieve dynamic risk prediction and assessment. Through classification, approval nodes are divided into high-risk approval nodes and normal approval nodes to achieve differentiated internal control. For high-risk approval nodes, the system will automatically insert new approval links and dynamically adjust their insertion frequency according to the risk level to ensure that high-risk operations are fully reviewed and controlled, which not only improves security in high-risk scenarios, but also avoids excessive intervention in low-risk situations, thereby improving approval efficiency.

[0061] The present invention adopts risk classification and internal control strengthening measures to effectively reduce the risk of approval links being bypassed, logical conflicts or process omissions, prevent fund leakage, duplicate payments and unauthorized operations, achieve optimal allocation of internal control resources, and enhance the risk management capabilities of enterprises in complex approval processes. Through real-time monitoring and dynamic adjustment, the transparency and compliance of the approval process are guaranteed, while the processing efficiency of the approval nodes is optimized, ensuring that enterprises maintain efficient operations while improving security, helping enterprises to respond to risk changes in a timely manner, and realizing dynamic optimization and continuous improvement of the internal control system, thereby ensuring financial and business security and providing strong support for the long-term stable operation of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0063] Figure 1The present invention is a method flow chart of an intelligent internal control test verification method.

[0064] Figure 2 The module diagram of an intelligent internal control test verification system of the present invention is shown in FIG. DETAILED DESCRIPTION

[0065] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0066] The present invention provides Figure 1 An intelligent internal control test verification method shown includes the following steps:

[0067] In the business scenario of multi-level approval process, the actual inspection information of all approval nodes is comprehensively collected to establish a complete analysis set;

[0068] The actual detection information includes the operation log, approval result, approval time, identity and authority of the approver of each approval node. These data reflect the actual operation of the approval process. The purpose of establishing the analysis set is to provide a detailed data basis for subsequent comparison and analysis, to ensure that all actual approval activities are accurately recorded, and to lay a solid foundation for identifying potential process loopholes and abnormal behaviors.

[0069] After each approval check of the approval node is completed, the approval nodes that are not executed as expected in the approval process are preliminarily identified, and the approval nodes that are preliminarily identified as not executed as expected are marked and filtered out;

[0070] The expected detection information is an ideal process formulated according to the company's internal control standards, business process design and approval policies, including the necessary approval nodes, sequence, authority and conditions.

[0071] After each approval test is completed, the approval nodes that are not executed as expected in the approval process are preliminarily identified, and the approval nodes that are preliminarily identified as not executed as expected are marked and screened out. The specific steps are as follows:

[0072] After the approval process is completed, the actual detection information of each approval node in the analysis set is compared with the expected detection information one by one, and the approval operations that do not meet expectations are found through the rule engine;

[0073] Unexpected approval operations, such as node skipping, over-authority approval, too short or too long execution time, and other abnormal behaviors. The comparison process needs to cover all approval nodes to ensure that each link is executed in accordance with the established rules and identify potential defects or anomalies in the process.

[0074] After the comparison is completed, the approval nodes that do not meet expectations are screened out;

[0075] Approval nodes that do not meet expectations are those abnormal nodes found during the comparison process. For example, a key node is not triggered as planned, or the approver completes the approval operation without authorization. By analyzing these abnormal situations, the system preliminarily identifies which nodes have the risk of bypassing logic or violating processes. This identification process also includes determining whether there are problems such as logical loops and redundant approvals. By classifying and recording the identified abnormal nodes, the system provides data support for subsequent internal control improvements and risk management, and indicates high-risk approval links that need to be focused on.

[0076] Once the approval nodes with abnormalities are initially identified, they will be marked and screened out to enter further analysis and processing flow;

[0077] Marking can take the form of labeling, assigning risk scores, or adding them to a dedicated exception handling queue to ensure that the system pays attention to these nodes in a timely manner.

[0078] For the approval nodes that are initially identified as not being executed as expected, the actual test information after the approval test is completed is further compared and analyzed with the expected test information, key features are extracted from the compared data, and the extracted features are analyzed and processed;

[0079] The analyzed and processed feature data is input into the pre-trained machine learning model, and the actual detection situation of the approval node is predicted through the machine learning model to assess the potential risk of repeated detection logic bypassing each approval node.

[0080] After each approval detection of the approval node is completed, key features are extracted from the compared data. The extracted key features include the frequency of the approval node not being triggered when it is expected to be executed and the degree of deviation between the actual processing time of the approval node and the expected processing time. After acquisition, the frequency of the approval node not being triggered when it is expected to be executed and the degree of deviation between the actual processing time of the approval node and the expected processing time are analyzed and processed, and an approval skip factor and an approval time anomaly factor are generated respectively. The approval skip factor is used to measure the degree of lack of integrity in the approval link, revealing the bypass risks or logical blind spots that may exist in the approval process. The approval time anomaly factor is used to evaluate the execution efficiency of the approval operation, revealing the abnormal acceleration or delay that may exist in the approval process.

[0081] After obtaining the approval skip factor and approval time abnormality factor generated by analyzing and processing the extracted key features, the approval skip factor and approval time abnormality factor are input into the pre-trained machine learning model to generate a detection logic bypass coefficient. The detection logic bypass coefficient is used to evaluate the potential risk of repeated detection logic bypassing each approval node.

[0082] A pre-trained machine learning model is a model that has been trained and optimized offline based on historical approval process data, feature analysis results, and known risk cases. The model has learned how to identify potential risk patterns from input data (such as approval skip factors and approval time anomaly factors). During the training phase, the system uses a large number of historical data samples that mark normal nodes and risky nodes, and continuously optimizes parameters based on model algorithms (such as logistic regression, random forest, neural network, etc.) to ensure that the model can accurately capture abnormal patterns in the approval process.

[0083] In this solution, the role of the model is to evaluate the logic bypass risk of each approval node. Specifically, when the characteristic data in the approval process (such as skip index, time anomaly index) is input into the model, the model will calculate a detection logic bypass coefficient, which quantifies the risk level of a certain approval node being bypassed or not executed as expected. Through advance training, the model has already mastered which characteristic patterns in the approval process are prone to risks. For example, nodes that are frequently skipped or nodes with significantly abnormal approval times are usually related to process bypass or logic conflicts. The model will make a comprehensive judgment based on these characteristics and output a risk coefficient for the enterprise to evaluate and process in real time.

[0084] The training process of the model not only relies on historical approval data, but also needs to be combined with the business characteristics of the enterprise and the logical relationship between the approval nodes. The training process usually includes the following steps: data preprocessing, feature selection, model training, model verification and optimization. In the preprocessing stage, the model will clean and standardize the data to eliminate noise and outliers. In the feature selection stage, key features such as approval skip factor and time anomaly index will be extracted, and the contribution of these features to the risk will be evaluated through data analysis tools. Then in the training stage, the labeled samples are input into the model, and the model parameters are continuously adjusted to improve the accuracy and generalization ability of the prediction. In the verification stage, the test set data is used to verify the model performance, and the model is optimized through methods such as cross-validation and confusion matrix analysis. The detection logic bypass coefficient generated by the model represents the risk level of the approval node being logically bypassed in actual operation. The detection logic bypass coefficient can quantify the risk level of the approval node in real time and provide enterprises with accurate decision-making basis.

[0085] If the approval node is not triggered frequently when it is expected to be executed, it indicates that there is a high potential risk of repeated detection logic bypassing each other at the approval node. If further comparison and analysis finds that a certain approval node is not triggered frequently when it is expected to be executed, it means that the node may be bypassed due to logic conflicts, overlapping authorizations, or lax process design. Frequently untriggered nodes reflect loopholes in the process, that is, the detection logic between multiple approval nodes may offset or cover each other, resulting in key links not being executed as required. This situation indicates that the control measures of the approval node may fail, and there are greater operational risks and compliance risks, especially in important processes such as large payments or fund approvals, which may cause serious problems such as fund leakage or fraud.

[0086] The specific steps for generating the approval skip factor after analyzing the frequency of approval nodes not being triggered when they are expected to be executed are as follows:

[0087] Collect the actual detection information and expected detection information of the approval node, and calculate the time priority coefficient of the approval node. The calculation expression is:

[0088]

[0089] , where P time (t) is the time priority coefficient of the approval node, which indicates the importance of the approval node in the process and is expressed as a weight function that changes over time. t is the execution time of the current approval node, t0 is the key time point of the approval node, such as a specific approval deadline or a high priority time window, and α is a parameter for adjusting the priority change rate. If α is large, the priority of the node rises sharply near the key time t0. If α is small, the priority changes more slowly.

[0090] The time priority coefficient reflects the importance of the approval node in different time periods. For example, the approval node is more critical during peak hours than at other times.

[0091] By analyzing the frequency and importance of approval nodes being skipped, the skip weight is calculated. The calculation expression of the skip weight is: W miss =R missed ·P time (t), where W miss The skip weight is the skip frequency and time priority of the approval node. If an approval node is frequently skipped at a critical moment, its skip weight will be higher. missed The skip frequency of the approval node is calculated as follows:

[0092]

[0093] , where N totalN is the total number of times the approval node is expected to be triggered. missed The number of times the approval node was not actually triggered, that is, the number of times the approval node should be executed but was not executed;

[0094] The approval skip factor is calculated by skipping weight to quantify the degree of completeness loss in the approval process. The calculation expression is:

[0095]

[0096] , where I skip is the approval skip factor, γ is the correction parameter, which is used to control the degree of correction of the approval skip factor by the non-skipped approval node;

[0097] The formula introduces a combination of exponential function and logarithmic function, so that when the skip weight is high, the approval skip factor shows a nonlinear growth; and when the approval node is not skipped frequently, the growth of the approval skip factor tends to be gentle.

[0098] It can be seen from the approval skip factor that for the approval nodes that are initially identified as not executed as expected, the larger the performance value of the approval skip factor generated after analyzing the frequency of the approval node being not triggered when it is expected to be executed, the higher the frequency of the approval node being not triggered when it is expected to be executed, indicating that the potential risk of duplicate detection logic bypassing each other in the approval node is greater. This means that there may be blind spots, overlapping authorizations, or conflicting conditions in the logic of the approval process, causing the approval node to be frequently bypassed or ignored. On the contrary, if the performance value of the approval skip factor is small, it means that the execution of the node is basically consistent with the expected requirements, the process integrity is high, and the risk of bypassing the duplicate detection logic is low.

[0099] For approval nodes that are initially identified as not being executed as expected, the system further compares the degree of deviation between the actual processing time and the expected processing time. If the deviation is large, it indicates that there is a high potential risk of repeated detection logic bypassing each other at this node. This time deviation may mean that the approval process is not executed at a normal pace. For example, if the approval time is too fast, it may be that the node is skipped or bypassed by the automated system; if the approval time is too long, it may indicate approval blockage, authority conflict, or human intervention. These anomalies usually indicate that the logic detection in the process fails, causing some nodes to be accidentally or intentionally bypassed, thereby creating risks. Therefore, nodes with high time deviation need to be further analyzed and reviewed to ensure the security and compliance of the approval process.

[0100] The specific steps for generating the approval time abnormality factor after analyzing the deviation between the actual processing time and the expected processing time of the approval node are as follows:

[0101] Define the time deviation function to quantify the difference between the actual processing time and the expected processing time of the approval node. The time deviation function serves as the basis for subsequent calculations. The function expression is: ΔT(t) = |T actual (t)-T expected , where ΔT(t) is the time deviation function, which represents the time error of the approval node at time t, T actual (t) is the actual processing time function of the approval node at time t, which describes the actual time consumed by the approval node at each time point. expected (t) is the expected processing time function of the approval node at time t. The expected processing time is the time that the approval node should take under the ideal state preset by the system;

[0102] The time error function represents the deviation between the actual operation time and the expected time of the approval node, calculated in absolute value form, to ensure that whether it is accelerated execution or delayed execution, the time deviation will be reflected. This is the basis for subsequent evaluation.

[0103] In order to evaluate the overall time deviation of the approval node, the time deviation function is integrated during the operation of the approval node to calculate the total accumulated time deviation. The calculation expression is:

[0104]

[0105] , where S(T) is the total time deviation accumulation, which means the total time deviation of the approval node in the entire operation time period, t 开始 is the start time of approval node processing, t 结束 The end processing time of the approval node;

[0106] By integrating the time deviation function, the total time deviation of the approval node in the entire processing process is calculated. The total time deviation reflects the overall time anomaly of the approval node. If the accumulated time deviation is large, it means that there is a large deviation in the execution process of the approval node.

[0107] After calculating the cumulative time deviation, the processing complexity of the approval node is further introduced to calculate the approval time anomaly factor to more comprehensively evaluate the efficiency of the approval operation. The calculation expression of the approval time anomaly factor is:

[0108]

[0109] , where Time Anomaly is the approval time abnormality factor, which is used to evaluate the abnormal degree of the execution efficiency of the approval node. C(t) is the processing complexity coefficient, which indicates the complexity of the approval node during the operation.

[0110]

[0111] , where D(τ) is the amount of data dynamically processed by the approval task at time τ (such as the size of the approval document, the number of data entries), W(τ) is the task complexity weight function, which indicates the complexity of the approval matter (such as the approval task involving multiple departments or complex contracts), P is the authority level of the approver, which is used to adjust the complexity benchmark (the higher the authority, the lower the complexity), τ represents a certain instantaneous time point in the integration process, and T expected is the total expected processing time of the approval node;

[0112] The setting of the task complexity weight function needs to be dynamically adjusted according to the specific attributes of the approval task and the complexity of the operation. Usually, it will take into account multiple factors such as the task type, data processing volume, the number of departments involved, the sensitivity of the approval matters, and the branches and levels of the operation process. For example, the task complexity weight of large-scale cross-departmental fund approval and complex contract review is greater than the approval of daily small reimbursements. The task complexity weight function can change over time. By setting weights at different time points (such as increased complexity during peak periods), it ensures that the system accurately reflects the complexity when the approval tasks change dynamically. The commonly used setting method is to set it based on the weighted average method of historical data;

[0113] This step generates the approval time anomaly factor by combining the accumulated time deviation with the processing complexity coefficient of the approval node. The introduction of processing complexity ensures that the index not only reflects the time deviation, but also takes into account the operational complexity of the approval node.

[0114] It can be seen from the approval time anomaly factor that for the approval nodes that are initially identified as not executed as expected, the larger the performance value of the approval time anomaly factor generated after analyzing the degree of deviation between the actual processing time of the approval node and the expected processing time, the higher the degree of deviation between the actual processing time of the approval node and the expected processing time, indicating that the potential risk of bypassing the duplicate detection logic of the approval node is greater. This may be due to abnormal operations caused by problems such as the approval link being skipped, the processing order being disrupted, or the approval process encountering logical conflicts. On the contrary, when the value of the approval time anomaly factor is small, it means that the actual processing time is relatively consistent with the expected processing time, and the execution efficiency of the approval node is high, indicating that the risk of logic bypassing of the approval node is small, and the process operation is more stable and compliant. Therefore, the size of the approval time anomaly factor can be used as an important reference indicator for identifying and evaluating node logic anomalies.

[0115] The machine learning model is not specifically limited here, and can realize the approval skipping factor I skip and approval time abnormal factor Time Anomaly Perform comprehensive analysis to generate detection logic bypass coefficients By LogicIn order to realize the technical solution of the present invention, the present invention provides a specific implementation method;

[0116] Detection logic bypass coefficient By Logic The resulting calculation formula is:

[0117]

[0118] , where p1 and p2 are the approval skipping factors I skip and approval time abnormal factor Time Anomaly The preset proportional coefficient, and p1 and p2 are both greater than 0.

[0119] It can be seen from the detection logic bypass coefficient that, for the approval nodes that are initially identified as not executed as expected, the larger the performance value of the approval skip factor generated after analyzing and processing the frequency of the approval nodes not being triggered when they are expected to be executed, the larger the performance value of the approval time anomaly factor generated after analyzing and processing the degree of deviation between the actual processing time of the approval node and the expected processing time, that is, the larger the performance value of the detection logic bypass coefficient generated by predicting the actual detection situation of the approval node through the pre-trained machine learning model, the greater the potential risk of repeated detection logic bypassing the approval node, and vice versa.

[0120] According to the prediction results of the machine learning model, the approval nodes are classified into high-risk approval nodes and normal approval nodes;

[0121] The detection logic bypass coefficient generated by predicting the actual detection situation of the approval node through the pre-trained machine learning model is compared with the pre-set detection logic bypass coefficient reference threshold to classify the approval nodes. The specific classification steps are as follows:

[0122] If the detection logic bypass coefficient is greater than the detection logic bypass coefficient reference threshold, the approval node is classified as a high-risk approval node. The high-risk approval node may have problems such as approval logic bypass and unauthorized operation.

[0123] If the detection logic bypass coefficient is less than or equal to the detection logic bypass coefficient reference threshold, the approval node is classified as a normal approval node, and the normal approval node is a node where the process execution is normal.

[0124] The purpose of classification is to take targeted control measures for approval nodes with different risk levels and improve the utilization efficiency of internal control resources.

[0125] For approval nodes classified as high-risk, take enhanced internal control measures and insert new approval links in high-risk approval nodes;

[0126] In order to ensure a balance between security and efficiency in the approval process, the insertion frequency of new approval links is dynamically adjusted according to the risk level of bypassing each other in duplicate detection logic of approval nodes;

[0127] The new approval steps inserted include:

[0128] Increase the approval level: Introduce higher-level management personnel to participate in the approval process to enhance the authority and rigor of the approval process.

[0129] Increase manual review: Have dedicated reviewers review the approval results to prevent loopholes in the automated process.

[0130] Multiple verification mechanism: Introduce dual or multiple authentication to ensure the authenticity and legality of approval operations.

[0131] These measures are designed to fill loopholes in the original approval process, prevent key control links from being bypassed, and ensure that high-risk operations are adequately reviewed and verified.

[0132] The insertion frequency of new approval links is dynamically adjusted based on the risk level of repeated detection logic bypassing each other between approval nodes. This dynamic adjustment mechanism can flexibly respond to changes in risks in the approval process, which not only ensures the security of the approval process, but also avoids unnecessary process redundancy and optimizes the implementation of internal control strategies.

[0133] According to the risk level of bypassing the duplicate detection logic of the approval nodes, the insertion frequency of the new approval link is dynamically adjusted. The specific steps are as follows:

[0134] Calculate the risk exceedance ratio of high-risk approval nodes. The calculation expression is:

[0135]

[0136] , where R exceed is the risk exceedance ratio, which is used to measure the degree to which the detection logic bypass coefficient of the approval node exceeds the detection logic bypass coefficient reference threshold. Logic is the detection logic bypass coefficient, which indicates the actual risk level of the approval node. threshold To detect the reference threshold of the logic bypass coefficient, it is set to the maximum bypass risk level that the enterprise considers acceptable;

[0137] After determining the risk exceedance ratio, the insertion frequency of the new approval link is dynamically adjusted according to the risk exceedance ratio. During regulation, a frequency adjustment factor is introduced to determine the insertion frequency of the new approval link. The calculation expression of the frequency adjustment factor is: Fadjust =ψ·ln(1+R exceed ), where F adjust is the frequency adjustment factor, which indicates the insertion frequency adjustment amount based on the actual risk level of the approval node. ψ is the adjustment coefficient, which depends on the enterprise risk tolerance and process complexity and is usually set to a constant (such as 1 or 0.5). exceed ) is used to smooth the risk exceedance ratio using the natural logarithm function to prevent the insertion frequency from growing too aggressively;

[0138] When the risk exceeds the ratio R exceed When it is higher, the frequency adjustment factor F adjust will increase, indicating that new approval links need to be inserted more frequently. The natural logarithm ln(1+R exceed ) makes the frequency adjustment process smoother and avoids excessive increase in approval links in extreme cases.

[0139] The new approval link insertion frequency is dynamically calculated based on the original approval link insertion frequency and the frequency adjustment factor. The calculation expression is: insert =f0·(1+F adjust ), where f insert is the new insertion frequency of the final application, indicating the actual insertion frequency of the new approval link under the current risk level, and f0 is the initial insertion frequency, indicating the insertion frequency of the approval link under the default setting;

[0140] The default frequency of approval link insertion refers to the frequency of new link insertion in the approval node when the system does not detect abnormal risks or bypass logic. This means that the basic control measures set by the enterprise in the regular process ensure that when there is no high risk, the system still performs a certain frequency of approval link insertion to maintain the basic internal control standards of the process. This frequency is usually set according to the company's internal control strategy, the complexity of the approval task, and the best practices in historical data. The design of the default insertion frequency needs to strike a balance between control costs and approval efficiency. It should neither make the approval process too cumbersome nor avoid potential loopholes due to the lack of necessary review links;

[0141] When the frequency adjustment factor F adjust As it increases, the final insertion frequency f insert will also increase, which means that the system will insert the approval link more frequently in high-risk scenarios. If the detection logic bypass coefficient is lower than the reference threshold, the frequency adjustment factor F adjust The insertion frequency will be close to 0 and maintained at the initial insertion frequency f0 to avoid unnecessary interference with the process.

[0142] Through the above-mentioned intelligent internal control test verification method, enterprises can accurately identify and control the risks in the approval process in a complex multi-level approval process, and achieve a balance between the safety and efficiency of the approval process. This method establishes a complete analysis set, comprehensively collects and compares actual detection information with expected detection information, and preliminarily identifies the approval nodes in the approval process that are not executed as expected. On this basis, the system extracts key features of the initially identified abnormal nodes, and inputs the extracted feature data into a pre-trained machine learning model to achieve dynamic risk prediction and assessment. The machine learning model can not only effectively detect the logical bypass risks between approval nodes, but also provide the ability to predict potential risks in the future, so that enterprises can take effective preventive measures before problems arise.

[0143] By classifying the approval nodes, the nodes are divided into high-risk approval nodes and normal approval nodes, thus achieving differentiated internal control. For high-risk approval nodes, the system will automatically insert new approval links and dynamically adjust the insertion frequency of approval links according to the risk level to ensure that high-risk operations are adequately reviewed and controlled. This dynamic control mechanism can not only improve security in high-risk scenarios, but also avoid excessive intervention in low-risk situations, thereby improving approval efficiency. At the same time, the system's risk classification and internal control strengthening measures effectively reduce the possibility of approval links being bypassed, logical conflicts or process omissions, and prevent fund leakage, duplicate payments and unauthorized operations.

[0144] This intelligent internal control test and verification solution achieves the optimal allocation of internal control resources and improves the risk management capabilities of enterprises in complex approval processes. Through real-time monitoring and dynamic adjustment, it not only ensures the transparency and compliance of the approval process, but also optimizes the processing efficiency of approval nodes, ensuring that enterprises maintain efficient operations while improving security. This solution enables enterprises to respond to risk changes in a timely manner and realize dynamic optimization and continuous improvement of internal control systems, thereby providing strong support for the long-term stable operation of enterprises on the basis of ensuring financial and business security.

[0145] The present invention provides Figure 2 An intelligent internal control test verification system shown includes a data collection module, an abnormal node identification module, a feature extraction and analysis module, a risk prediction module, a risk classification module, an internal control strengthening module and a dynamic control module;

[0146] The data collection module comprehensively collects the actual inspection information of all approval nodes in the business scenario of multi-level approval process and establishes a complete analysis set;

[0147] Abnormal node identification module: After each approval node inspection is completed, it preliminarily identifies the approval nodes that are not executed as expected in the approval process, and marks and screens out the approval nodes that are preliminarily identified as not executed as expected;

[0148] The feature extraction and analysis module further compares and analyzes the actual detection information after the approval detection is completed with the expected detection information for the approval nodes that are initially identified as not executed as expected, extracts key features from the compared data, and analyzes and processes the extracted features;

[0149] The risk prediction module inputs the analyzed and processed feature data into a pre-trained machine learning model, uses the machine learning model to predict the risk of the actual detection situation of the approval node, and evaluates the potential risk of repeated detection logic bypassing each approval node;

[0150] The risk classification module classifies approval nodes according to the prediction results of the machine learning model, dividing the approval nodes into high-risk approval nodes and normal approval nodes;

[0151] The internal control strengthening module takes strengthened internal control measures for approval nodes classified as high-risk and inserts new approval links into high-risk approval nodes;

[0152] Dynamic control module, in order to ensure the balance between security and efficiency of the approval process, dynamically adjusts the insertion frequency of new approval links according to the risk level of bypassing each other in the duplicate detection logic of the approval nodes;

[0153] An intelligent internal control test verification method provided by an embodiment of the present invention is implemented by the above-mentioned intelligent internal control test verification system. The specific methods and processes of an intelligent internal control test verification system are detailed in the embodiment of the above-mentioned intelligent internal control test verification method, which will not be repeated here.

[0154] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0155] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0156] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An intelligent internal control test verification method, characterized in that: The following steps are involved: In the business scenario of multi-level approval process, the actual inspection information of all approval nodes is comprehensively collected to establish a complete analysis set; After each approval check of the approval node is completed, the approval nodes that are not executed as expected in the approval process are preliminarily identified, and the approval nodes that are preliminarily identified as not executed as expected are marked and filtered out; For the approval nodes that are initially identified as not being executed as expected, the actual test information after the approval test is completed is further compared and analyzed with the expected test information, key features are extracted from the compared data, and the extracted features are analyzed and processed; The analyzed and processed feature data is input into the pre-trained machine learning model, and the actual detection situation of the approval node is predicted through the machine learning model to assess the potential risk of repeated detection logic bypassing each approval node. According to the prediction results of the machine learning model, the approval nodes are classified into high-risk approval nodes and normal approval nodes; For approval nodes classified as high-risk, take enhanced internal control measures and insert new approval links in high-risk approval nodes; In order to ensure a balance between security and efficiency in the approval process, the insertion frequency of new approval links is dynamically adjusted according to the risk level of bypassing duplicate detection logic of approval nodes.

2. The intelligent internal control test verification method according to claim 1, characterized in that: After each approval test is completed, the approval nodes that are not executed as expected in the approval process are preliminarily identified, and the approval nodes that are preliminarily identified as not executed as expected are marked and screened out. The specific steps are as follows: After the approval process is completed, the actual detection information of each approval node in the analysis set is compared with the expected detection information one by one, and the approval operations that do not meet expectations are found through the rule engine; After the comparison is completed, the approval nodes that do not meet expectations are screened out; Once the approval nodes with abnormalities are initially identified, they will be marked and screened out to enter further analysis and processing flow.

3. The intelligent internal control test verification method according to claim 1, characterized in that: After each approval detection of the approval node is completed, key features are extracted from the compared data. The extracted key features include the frequency of the approval node not being triggered when it is expected to be executed and the degree of deviation between the actual processing time of the approval node and the expected processing time. After acquisition, the frequency of the approval node not being triggered when it is expected to be executed and the degree of deviation between the actual processing time of the approval node and the expected processing time are analyzed and processed to generate the approval skip factor and the approval time abnormal factor respectively. The degree of integrity loss of the approval link is measured by the approval skip factor, and the execution efficiency of the approval operation is evaluated by the approval time abnormal factor.

4. The intelligent internal control test verification method according to claim 3, characterized in that: After obtaining the approval skip factor and approval time abnormality factor generated by analyzing and processing the extracted key features, the approval skip factor and approval time abnormality factor are input into the pre-trained machine learning model to generate a detection logic bypass coefficient. The detection logic bypass coefficient is used to evaluate the potential risk of repeated detection logic bypassing each approval node.

5. The intelligent internal control test verification method according to claim 4, characterized in that: The detection logic bypass coefficient generated by predicting the actual detection situation of the approval node through the pre-trained machine learning model is compared with the pre-set detection logic bypass coefficient reference threshold to classify the approval nodes. The specific classification steps are as follows: If the detection logic bypass coefficient is greater than the detection logic bypass coefficient reference threshold, the approval node is classified as a high-risk approval node; If the detection logic bypass coefficient is less than or equal to the detection logic bypass coefficient reference threshold, the approval node is classified as a normal approval node, and the normal approval node is a node where the process execution is normal.

6. The intelligent internal control test verification method according to claim 3, characterized in that: The specific steps for generating the approval skip factor after analyzing the frequency of approval nodes not being triggered when they are expected to be executed are as follows: Collect the actual detection information and expected detection information of the approval node, and calculate the time priority coefficient of the approval node. The calculation expression is: Where P time (t) is the time priority coefficient of the approval node, indicating the importance of the approval node in the process, t is the execution time of the current approval node, t0 is the key time point of the approval node, and α is the parameter for adjusting the priority change rate; By analyzing the frequency and importance of approval nodes being skipped, the skip weight is calculated. The calculation expression of the skip weight is: W miss =R missed ·P time (t), where W miss is the skip weight, which comprehensively considers the skip frequency and time priority of the approval node. missed The skip frequency of the approval node is calculated as follows: Where N total N is the total number of times the approval node is expected to be triggered. missed The number of times the approval node was not actually triggered, that is, the number of times the approval node should be executed but was not executed; The approval skip factor is calculated by skipping weight to quantify the degree of completeness loss in the approval process. The calculation expression is: In the formula, I skip is the approval skip factor, and γ is the correction parameter, which is used to control the degree of correction of the approval skip factor by the non-skipped approval node.

7. The intelligent internal control test verification method according to claim 3, characterized in that: The specific steps for generating the approval time abnormality factor after analyzing the deviation between the actual processing time and the expected processing time of the approval node are as follows: Define the time deviation function to quantify the difference between the actual processing time and the expected processing time of the approval node. The time deviation function serves as the basis for subsequent calculations. The function expression is: ΔT(t) = |T actual (t)-T expected (t)|, where ΔT(t) is the time deviation function, which represents the time error of the approval node at time t, T actual (t) is the actual processing time function of the approval node at time t, T expected (t) is the expected processing time function at the approval node at time t; In order to evaluate the overall time deviation of the approval node, the time deviation function is integrated during the operation of the approval node to calculate the total accumulated time deviation. The calculation expression is: In the formula, S(T) is the total time deviation accumulation, which means the total time deviation of the approval node in the entire operation time period, t 开始 is the start time of approval node processing, t 结束 The end processing time of the approval node; After calculating the cumulative time deviation, the processing complexity of the approval node is further introduced to calculate the approval time anomaly factor to more comprehensively evaluate the efficiency of the approval operation. The calculation expression of the approval time anomaly factor is: Where Time Anomaly is the approval time abnormality factor, which is used to evaluate the abnormal degree of the execution efficiency of the approval node. C(t) is the processing complexity coefficient, which indicates the complexity of the approval node during the operation. Where D(τ) is the amount of data dynamically processed by the approval task at time τ, W(τ) is the task complexity weight function, which indicates the complexity of the approval matter, P is the authority level of the approver, which is used as the basis for adjusting the complexity, τ represents a certain instantaneous time point in the integration process, and T expected The total expected processing time of the approval node.

8. The intelligent internal control test verification method according to claim 5, characterized in that: According to the risk level of bypassing the duplicate detection logic of the approval nodes, the insertion frequency of the new approval link is dynamically adjusted. The specific steps are as follows: Calculate the risk exceedance ratio of high-risk approval nodes. The calculation expression is: In the formula, R exceed is the risk exceedance ratio, which is used to measure the degree to which the detection logic bypass coefficient of the approval node exceeds the detection logic bypass coefficient reference threshold. Logic is the detection logic bypass coefficient, which indicates the actual risk level of the approval node. threshold Bypass coefficient reference threshold for detection logic; After determining the risk exceedance ratio, the insertion frequency of the new approval link is dynamically adjusted according to the risk exceedance ratio. During regulation, a frequency adjustment factor is introduced to determine the insertion frequency of the new approval link. The calculation expression of the frequency adjustment factor is: F adjust =ψ·ln(1+R exceed ), where F adjust is the frequency adjustment factor, which represents the insertion frequency adjustment amount based on the actual risk level of the approval node, and ψ is the adjustment coefficient; The new approval link insertion frequency is dynamically calculated based on the original approval link insertion frequency and the frequency adjustment factor. The calculation expression is: insert =f0·(1+F adjust ), where f insert is the new insertion frequency of the final application, indicating the actual insertion frequency of the new approval link under the current risk level, and f0 is the initial insertion frequency, indicating the insertion frequency of the approval link under the default condition.

9. An intelligent internal control test verification system, used to implement the intelligent internal control test verification method according to any one of claims 1 to 8, characterized in that: It includes data collection module, abnormal node identification module, feature extraction and analysis module, risk prediction module, risk classification module, internal control strengthening module and dynamic control module; The data collection module comprehensively collects the actual inspection information of all approval nodes in the business scenario of multi-level approval process and establishes a complete analysis set; Abnormal node identification module: After each approval node inspection is completed, it preliminarily identifies the approval nodes that are not executed as expected in the approval process, and marks and screens out the approval nodes that are preliminarily identified as not executed as expected; The feature extraction and analysis module further compares and analyzes the actual detection information after the approval detection is completed with the expected detection information for the approval nodes that are initially identified as not executed as expected, extracts key features from the compared data, and analyzes and processes the extracted features; The risk prediction module inputs the analyzed and processed feature data into a pre-trained machine learning model, uses the machine learning model to predict the risk of the actual detection situation of the approval node, and evaluates the potential risk of repeated detection logic bypassing each approval node; The risk classification module classifies approval nodes according to the prediction results of the machine learning model, dividing the approval nodes into high-risk approval nodes and normal approval nodes; The internal control strengthening module takes strengthened internal control measures for approval nodes classified as high-risk and inserts new approval links into high-risk approval nodes; The dynamic control module, in order to ensure the balance between security and efficiency in the approval process, dynamically adjusts the insertion frequency of new approval links according to the risk level of bypassing each other in the duplicate detection logic of approval nodes.

Images