Administrative institution internal control process optimization and supervision method and system based on AI

Abstract

The invention discloses an AI-based method and system for optimizing and supervising an internal control process of an administrative institution, and relates to the technical field of intelligent management. Time sequence message transmission coding is carried out on the benchmark process and the actual process to achieve deep semantic alignment of the benchmark path and the actual path, the process execution semantic offset coefficient between the benchmark process and the actual process after semantic alignment is further quantified, and the process execution semantic offset coefficient is compared with a preset threshold value. Determining whether a path offset early warning prompt is generated or not; according to the method and the device, the process deviation can be captured in real time, early warning can be triggered, the real-time performance and objectivity of supervision are enhanced, abnormal risks in process execution are effectively prevented through an intelligent early warning mechanism, and the dynamic management and control capability with flexibility and normativity is provided for administrative institutions.

Description

Technical Field

[0001] This application relates to the field of intelligent management technology, and more specifically, to an AI-based method and system for optimizing and supervising the internal control process of administrative institutions and public institutions. Background Art

[0002] In the optimization and supervision of the internal control process of administrative institutions and public institutions, there are many defects and deficiencies in the traditional process execution monitoring methods. First of all, the current supervision methods mainly rely on means such as manual inspection, paper approval, and regular reports, with low efficiency and difficulty in adapting to the complex and changeable business needs of modern administrative institutions and public institutions. This traditional manual supervision method is easily affected by subjective factors, resulting in an increased risk of omissions and biases, and unable to comprehensively and accurately reflect the actual execution of the process.

[0003] In addition, although some units introduce basic information tools, their functions are mostly limited to post-event recording and simple statistics, unable to achieve real-time tracking of the process execution path, and the existing early warning mechanism overly relies on manual experience, lacking the intelligent analysis ability for dynamic association of time-series data, and it is difficult to identify process deviation and potential risks in a timely manner. These problems together lead to limited supervision effectiveness and lagged intervention in violations, making it difficult to meet the urgent needs of modern administrative institutions and public institutions for dynamic and intelligent internal control management.

[0004] Therefore, an optimized AI-based method for optimizing and supervising the internal control process of administrative institutions and public institutions is needed to solve the above technical problems. Summary of the Invention

[0005] In order to solve the above technical problems, this application is proposed.

[0006] According to one aspect of this application, an AI-based method for optimizing and supervising the internal control process of administrative institutions and public institutions is provided, which includes:

[0007] Set a benchmark process;

[0008] Perform semantic embedding encoding on each key node benchmark description in the benchmark process to obtain a time queue of benchmark process key node semantic embedding encoding features;

[0009] Obtain the actual process execution path, and perform structured encoding on each actual process execution key data in the actual process execution path to obtain a time queue of actual process execution key node encoding features;

[0010] Perform time-series message passing encoding on the time queue of actual process execution key node encoding features and the time queue of benchmark process key node semantic embedding encoding features to obtain actual process execution path semantic encoding features and benchmark process semantic embedding encoding features;

[0011] Determine whether to generate a path deviation warning prompt based on the process execution semantic deviation degree between the semantic encoding features of the actual process execution path and the semantic embedding encoding features of the benchmark process.

[0012] According to another aspect of the present application, there is provided an AI-based internal control process optimization and supervision system for administrative institutions, which includes:

[0013] A benchmark setting module for setting a benchmark process;

[0014] A benchmark process encoding module for performing semantic embedding encoding on the benchmark descriptions of each key node in the benchmark process to obtain a time queue of semantic embedding encoding features of the benchmark process key nodes;

[0015] An actual process encoding module for obtaining the actual process execution path and performing structured encoding on each actual process execution key data in the actual process execution path to obtain a time queue of actual process execution key node encoding features;

[0016] A time series message passing encoding module for performing time series message passing encoding on the time queue of actual process execution key node encoding features and the time queue of semantic embedding encoding features of benchmark process key nodes to obtain actual process execution path semantic encoding features and benchmark process semantic embedding encoding features;

[0017] A warning prompt module for determining whether to generate a path deviation warning prompt based on the process execution semantic deviation degree between the actual process execution path semantic encoding features and the benchmark process semantic embedding encoding features.

[0018] Beneficial effects: Compared with the prior art, the present application sets a benchmark process as a standard template, combines the real-time data flow of the actual business system, and performs time series message passing encoding on the benchmark process and the actual process respectively to achieve deep semantic alignment between the benchmark and the actual path. Further, it quantifies the process execution semantic deviation coefficient between the benchmark process and the actual process after semantic alignment, and determines whether to generate a path deviation warning prompt based on the comparison between the process execution semantic deviation coefficient and a preset threshold, which can capture process deviations in real time and trigger warnings, enhancing the real-time and objectivity of supervision. At the same time, through the intelligent warning mechanism, it effectively prevents abnormal risks in process execution, providing administrative institutions with dynamic control capabilities that are both flexible and standardized. Description of the Drawings

[0019] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail with reference to the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 It is a flowchart of an AI-based internal control process optimization and supervision method for administrative institutions according to an embodiment of the present application;

[0021] Figure 2 It is a schematic diagram of data flow of an AI-based internal control process optimization and supervision method for administrative institutions according to an embodiment of the present application;

[0022] Figure 3 A block diagram of an AI-based internal control process optimization and supervision system for administrative institutions according to an embodiment of the present application. Detailed implementation manners

[0023] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0024] As shown in the present application and the claims, unless the context clearly indicates an exception, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0025] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0026] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be executed precisely in sequence. On the contrary, as needed, various steps can be processed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps can be removed from these processes.

[0027] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0028] It should be noted that all the acquisition and processing of information or data in the present application are carried out on the premise of complying with the corresponding data protection regulations and policies and obtaining authorization from the authority manager.

[0029] In the technical solution of the present application, an AI-based internal control process optimization and supervision method for administrative institutions is proposed. Figure 1 FIG. is a flowchart of an AI-based internal control process optimization and supervision method for administrative institutions according to an embodiment of the present application; Figure 2 FIG. is a schematic diagram of data flow of an AI-based internal control process optimization and supervision method for administrative institutions according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the AI-based internal control process optimization and supervision method for administrative institutions according to the embodiment of the present application includes the steps: S1, setting a benchmark process; S2, performing semantic embedding encoding on the benchmark descriptions of each key node in the benchmark process to obtain a time queue of benchmark process key node semantic embedding encoding features; S3, obtaining the actual process execution path, and performing structured encoding on each actual process execution key data in the actual process execution path to obtain a time queue of actual process execution key node encoding features; S4, performing sequential message passing encoding on the time queue of actual process execution key node encoding features and the time queue of benchmark process key node semantic embedding encoding features to obtain actual process execution path semantic encoding features and benchmark process semantic embedding encoding features; S5, determining whether to generate a path deviation warning prompt based on the process execution semantic deviation degree between the actual process execution path semantic encoding features and the benchmark process semantic embedding encoding features.

[0030] Specifically, in S1, a benchmark process is set. Specifically, due to the lack of a unified standard in the traditional management mode, the process execution has a high degree of randomness and the supervision basis is fuzzy. The benchmark process transforms the ideal business specifications into digital templates through process twin technology, clarifying the power and responsibility boundaries, execution order, and compliance standards of each link. In the technical solution of the present application, through the mapping of the benchmark process and real-time data, the subjective deviation under the dependence on artificial experience is eliminated, ensuring the objectivity and consistency of the supervision standard.

[0031] Specifically, in S2, semantic embedding encoding is performed on the benchmark descriptions of each key node in the benchmark process to obtain a time queue of semantic embedding encoding features of the key nodes in the benchmark process. Specifically, in the traditional supervision system, although the process rules defined in the institutional text (such as "budget approval requires preliminary review by the department, review by the finance department, and final review by the management in charge") have clear operation guidelines, their semantic connotations have long existed in the form of discrete texts and are difficult to be directly parsed by machines into quantifiable logical relationships. In particular, the Bert model, with its bidirectional attention mechanism and deep semantic representation ability, can penetrate the surface grammar structure of the node description text and capture the deep semantic associations of key nodes in dimensions such as approval authority, business logic, and compliance boundaries. Therefore, in the technical solution of this application, the benchmark descriptions of each key node in the benchmark process are input into the benchmark process node semantic encoder based on the Bert model to obtain a time queue of benchmark process key node semantic embedding encoding vectors as the time queue of semantic embedding encoding features of the key nodes in the benchmark process. It is worth mentioning that through the semantic encoding of each key node by Bert, the originally scattered institutional clauses are transformed into a vector sequence with temporal correlation. This sequence not only retains the execution order between nodes (such as the temporal dependence of preliminary review → review → final review), but also implicitly expresses the intensity of logical constraints between nodes through the vector distance in the embedding space (such as the compliance correlation between "finance review" and "audit reporting"). This digital expression breaks through the limitations of the traditional rule engine relying on hard-coded logic, enabling the semantic features of the benchmark process to adapt to the flexible interpretation requirements of ambiguous descriptions in institutional texts. For example, when the system stipulates that "special matters can shorten the approval time limit", the Bert model can dynamically adjust the semantic weights of the corresponding nodes according to the context instead of rigidly executing fixed thresholds. In this way, a high-dimensional semantic reference system is provided for the subsequent real-time comparison of the actual process, improving the forward-looking warning ability for process anomalies.

[0032] Specifically, in step S3, obtain the actual process execution path, and perform structured encoding on each piece of key data in the actual process execution path to obtain a time queue of key node encoding features of the actual process execution. Specifically, in the embodiments of the present application, first, receive the key data of the actual process execution from the actual business system to obtain an actual process execution path composed of a time queue of the key data of the actual process execution. Since the execution traces of the actual process are scattered in heterogeneous systems such as finance, approval, and archives, and there are differences in data formats and recording standards among different departments, it is difficult for traditional manual summarization to capture the complete process execution timing logic. By extracting key data and constructing a time queue, in essence, fragmented operation records are reconstructed into an execution path map with causal associations. The time queue of the key data of the actual process execution can map the real rhythm of the process advancement (such as abnormal retention or skipping execution of a certain link) by retaining the sequence and interval duration of node execution, providing a process timing context for semantic deviation analysis; in addition, the structured integration of key data enables the system to identify hidden violation patterns across departments and systems (such as reverse operations of "executing first and then supplementing approval"), rather than only focusing on the compliance of a single node. This transformation from discrete data to a coherent path lays a dynamic data foundation for the virtual-real comparison of the process twin, enabling the monitoring mechanism to penetrate the business appearance and capture hidden logical faults and compliance risks in the process execution.

[0033] Then, considering that the key data in the actual process execution path (such as approval opinions, operation logs, form fields) often presents a fragmented and unstructured form, and there are significant differences in the expression forms and semantic granularities of the data generated by different business systems (for example, the description differences of "payment review" in the financial system and "contract acceptance" in the procurement system in the operation records). This data heterogeneity makes it difficult for traditional supervision methods to directly extract the process logic associations across systems, and it is even more impossible to effectively compare with the semantic features of the benchmark process. Therefore, in the technical solution of the present application, use the actual process embedding matrix to perform structured encoding on each piece of key data in the actual process execution path to obtain a time queue of key node encoding vectors of the actual process execution as the time queue of key node encoding features of the actual process execution. Through the pre-trained actual process embedding matrix, the system can transform scattered business operations into a unified vector space expression. This structured encoding enables the originally isolated key node data to form a semantic chain with context awareness ability in the time queue, providing a feature base with machine-parsable logical relationships for subsequent timing message passing analysis.

[0034] In particular, the S4 performs time-series message transmission encoding on the time queue of the encoding features of the key nodes of the actual process execution and the time queue of the semantic embedding encoding features of the key nodes of the benchmark process to obtain the semantic encoding features of the actual process execution path and the semantic embedding encoding features of the benchmark process. Specifically, the operation of the actual business process is essentially a complex system coupled in time and space. The execution order between nodes (such as project approval must precede budget allocation) not only reflects the causal law in the time dimension, but also implies the topological relationship of rights and responsibilities in the departmental collaboration network (such as the spatial position of the financial review node determines its risk interception ability). However, the existing technologies mostly use isolated time series analysis or static rule matching, which is difficult to model the dynamic context in the process advancement (such as how the delay of a certain link affects the subsequent nodes), and it is also impossible to quantify the spatial influence of the node in the organizational structure (such as the transmission effect of the abnormality of the financial department node on the overall process). Therefore, in order to break through the limitations of static feature comparison and model the semantic evolution law of the process execution path in the dimension of time-space fusion, in the technical solution of the present application, the time queue of the encoding features of the key nodes of the actual process execution and the time queue of the semantic embedding encoding features of the key nodes of the benchmark process are encoded by time-series message passing to obtain the semantic encoding features of the actual process execution path and the semantic embedding encoding features of the benchmark process. Specifically, in the technical solution of the present application, the temporal confidence (such as operation timeliness) and spatial confidence (such as departmental responsibility weight) of the process nodes are jointly modeled by using the time-space collaborative constraint factors, so as to construct the semantic encoding features of the actual process execution path and the semantic embedding encoding features of the benchmark process that are both sensitive to time evolution and aware of structural dependence. In particular, in this process, taking the time queue of the coding features of the key nodes of the actual process execution as an example, through the calculation of the spatiotemporal coordination constraint factors of the actual process transmission, the system can identify the abnormal retention of key nodes (such as the time confidence decay triggered by the timeout of a certain approval link) or unauthorized operation (such as the spatial confidence shift caused by the intervention of non-competent departments); on the other hand, the message transmission structure modulation mechanism simulates the selectivity (such as the enhanced transmission of important node information) and attenuation (such as the weakening of redundant operation information) of information transmission in the actual business process by dynamically adjusting the propagation weight of the coding vector. The semantic embedding features of the benchmark process use the same mechanism to transform the "ideal spatiotemporal relationship" implied in the institutional norms (such as the departmental collaboration density that the compliance process should have) into a computable feature vector, thereby establishing a multi-dimensional reference system for dynamic comparison. This organic propagation mechanism enables the semantic coding features to not only retain the linear temporal trajectory of process execution, but also embed the implicit rule network in the organizational structure (such as the authority gradient of cross-level approval), ultimately achieving the leap from "mechanical comparison" to "semantic deduction". For example, when the actual process leads to implicit violations due to broken departmental collaboration or disordered temporal logic, the system can accurately locate the deep-seated causes of compliance issues through semantic offset detection under the constraints of spatiotemporal coordination, rather than just capturing surface behavioral deviations.

[0035] Specifically, taking the time queue of the key node coding features of the actual process execution as an example, the specific steps for performing timing message passing encoding on the time queue of the key node coding features of the actual process execution are as follows: First, calculate the process node spatio-temporal coordination confidence constraint factor of the time queue of the key node coding features of the actual process execution to construct the actual process transfer spatio-temporal coordination constraint factor. Specifically, in the embodiments of the present application, first, input the time queue of the key node coding vectors of the actual process execution into a process node sequence encoder based on a recurrent neural network to obtain a time series of the initial coding vectors of the actual process sequence transfer. Specifically, the key nodes of the actual business process (such as "department preliminary review - financial review - management final review" in budget approval) are not isolated events, and their execution order and interval duration form a causal sequence chain. However, traditional monitoring mechanisms based on static rules or discrete event triggers are difficult to capture the dynamic dependency relationships between nodes. Through the recurrent connection structure of the RNN, the system can break through the cognitive limitation of simple sorting in the time dimension in traditional supervision, and model implicit timing patterns such as approval rhythm and node intervals during the process of information transfer across time steps, making cross-node impact chains such as "abnormal delay in the financial review link leading to backlog in subsequent audit processes" become explicit. During this process, the RNN, through its inherent timing memory ability, transforms the time queue of the key node coding features of the actual process execution into a time series of the initial coding vectors of the actual process sequence transfer with context awareness. Each initial coding vector of the actual process sequence transfer not only contains its own semantic features, but also carries the execution trajectory information of the previous node through the hidden state (such as the coding vector of the "management final review" node implies the rigor of the previous review link). The accumulation and transfer of this timing information provide a feature base rich in dynamic associations for the subsequent calculation of the spatio-temporal coordination constraint factor, enabling the system to distinguish timing patterns that are seemingly compliant but actually abnormal (such as unauthorized temporary nodes interspersed in compliant steps). This deep timing modeling ability enables the supervision system to break through the mechanical verification of the surface integrity of the process and turn to the intelligent perception of the internal rhythm of the process operation, providing a feature expression with dynamic association characteristics for subsequent spatio-temporal coordination analysis. In a specific example, input the time queue of the key node coding vectors of the actual process execution into a process node sequence encoder based on a recurrent neural network according to the following coding formula to obtain a time series of the initial coding vectors of the actual process sequence transfer; where the coding formula is:

[0036]

[0037]

[0038] Wherein, is the time queue of the key node coding vectors of the actual process execution, They are the 1st, 2nd, rd, and th actual process execution key node encoding vectors in the time queue of the actual process execution key node encoding vectors respectively, recurrent neural network, They are the 1st, 2nd, rd, and th actual process sequence passing initial encoding vectors in the time series of the actual process sequence passing initial encoding vectors respectively.

[0039] Next, calculate the actual process time series confidence constraint factors of each actual process sequence passing the initial encoding vector in the time series of the actual process sequence passing the initial encoding vector. Specifically, in the execution trajectory of the actual business process, there are significant differences in the importance of the time dimension of different nodes: the time compliance of some key nodes (such as the joint review and signature before large - amount fund payment) directly affects the overall risk level, while the time - series fluctuations of conventional nodes (such as the interval duration of general document circulation) may be within a reasonable elastic range. Traditional monitoring mechanisms based on uniform weights or manual experience - based weighting are difficult to dynamically identify the value gradient of time - series features, resulting in insufficient sensitivity of the risk warning system to substantial time - series violations. By introducing the time - series confidence constraint factor, the system can break through the limitation of static weight allocation and establish a dynamic mapping relationship of time sensitivity in the business process. In this process, the system not only considers the time - series attributes of the nodes themselves (such as the legal time limit of the approval link), but also can dynamically adjust the confidence weights in combination with the context semantic environment (such as the historical compliance record of the special approval process for urgent matters). This dynamic weighting mechanism enables time - series analysis to shift from rigid rule matching to elastic value evaluation, providing a refined weight basis for the subsequent integration of spatio - temporal collaborative constraint factors. In a specific example, calculate the actual process time series confidence constraint factors of each actual process sequence passing the initial encoding vector in the time series of the actual process sequence passing the initial encoding vector with the following calculation formula; where the calculation formula is:

[0040]

[0041] Where, function, and respectively represent trainable weighted hyperparameters, and are weight matrices, represents vector multiplication, is the actual process execution key node weight vector, is the time queue of the actual process time - series confidence measurement factors, function, is the actual process time - series confidence constraint factor.

[0042] Subsequently, calculate the actual process space confidence constraint factor of each actual process sequence transmitting the initial coding vector in the time series of the actual process sequence transmitting the initial coding vector. Specifically, the operation of the actual business process not only involves the chronological order in the time dimension, but also implies complex spatial topological relationships: for example, in the procurement process, the spatial position difference between the "requisition department submission" node and the "financial review" node determines the different risk conduction paths (the former is at the starting point of the process, and the latter is at the key risk interception point). Traditional monitoring mechanisms based on temporal features are difficult to quantify the impact of such spatial structure differences on process compliance. By introducing the spatial confidence constraint factor, the system can break through the observational limitation of a single time dimension and deconstruct the hierarchical structure importance of nodes in the business process. In this process, the system not only analyzes the explicit functional attributes of the nodes (such as the approval authority level), but also implicitly captures the implicit restrictive relationships between the nodes through the vector distance embedded in the space (such as the scope of the supervision effect of the audit node on the previous links). This dynamic weighting mechanism transforms the static organizational structure of the business process into a computable spatial influence map, providing a structured weight basis for subsequent spatio-temporal collaborative analysis. In a specific example, calculate the actual process space confidence constraint factor of each actual process sequence transmitting the initial coding vector in the time series of the actual process sequence transmitting the initial coding vector with the following calculation formula; where the calculation formula is:

[0043]

[0044] Wherein, Exponential operation, Represents the square of the norm, Represents The corresponding spatial confidence score value, Is the actual process space confidence constraint factor.

[0045] Then, based on the actual process space confidence constraint factor and the actual process timing confidence constraint factor of each actual process sequence for transmitting the initial coding vector, construct the actual process transmission spatio-temporal coordination constraint factor of each actual process sequence for transmitting the initial coding vector. Specifically, first, fuse the actual process space confidence constraint factor and the actual process timing confidence constraint factor to obtain the initial actual process transmission spatio-temporal coordination constraint factor. Specifically, the compliance judgment of the actual business process not only depends on the timing compliance of node execution (such as the time limit requirement for the approval link), but also needs to consider its spatial power and responsibility positioning in the organizational structure (such as the functional boundary of cross-departmental nodes). The traditional monitoring mechanism based on a single dimension (time or space) is difficult to cope with complex scenarios. By constructing the spatio-temporal coordination constraint factor, the system can break through the analysis limitation of dimension fragmentation and model the real state of process execution from a fused perspective. In this process, by non-linearly fusing the timing confidence constraint factor (reflecting the time sensitivity of nodes) and the space confidence constraint factor (characterizing the structural importance of nodes), the system can adaptively generate weight parameters reflecting the spatio-temporal comprehensive value of process nodes. This fusion mechanism not only considers the feature intensity of a single dimension, but also captures the composite effect generated by spatio-temporal interaction (such as the exponential growth of risk caused by the time delay of key spatial nodes), laying a foundation for the accurate calculation of process deviation. In a specific example, the following fusion formula is used to fuse the actual process space confidence constraint factor and the actual process timing confidence constraint factor to obtain the initial actual process transmission spatio-temporal coordination constraint factor; where, the fusion formula is:

[0046]

[0047] Wherein, function, and are fusion weight parameters, is the initial actual process transmission spatio-temporal coordination constraint factor.

[0048] Then, the initial actual process transfer spatiotemporal coordination constraint factor is optimized by spatiotemporal curvature compensation to obtain the actual process transfer spatiotemporal coordination constraint factor. In particular, the spatiotemporal coupling characteristics of the actual business process (such as the timing dependence and the topological relationship of rights and responsibilities in cross-departmental collaboration) can be regarded as dynamic trajectories in a high-dimensional manifold space in mathematical representation, and simple linear fusion (such as weighted summation) is prone to the negative attraction effect of the single-dimensional attention mechanism (such as excessive amplification of the time dimension abnormality leading to distortion of the spatial structure), causing non-Euclidean geometric distortion of the fusion space. For example, in the budget approval process, if the emergency special approval feature (high temporal confidence constraint) of the time dimension is directly superimposed with the leapfrog approval feature (low spatial confidence constraint) of the space dimension, it may cause the fusion space to have a "saddle-shaped" curvature, causing the compliance judgment to deviate from the real semantic logic, and misjudge "compliance special approval" as "illegal leapfrogging". Therefore, in a preferred example of the present application, the initial actual process transfer spatiotemporal coordination constraint factor is optimized by spatiotemporal curvature compensation to obtain the actual process transfer spatiotemporal coordination constraint factor. Specifically, the curvature compensation mechanism is used to eliminate the nonlinear deviation caused by the single-mode attention mechanism (such as focusing only on a single dimension of time or space), and pull the distribution of the spatiotemporal synergy constraint factor from the non-planar manifold back to the Euclidean space, thereby maintaining the geometric flatness of the fusion space. Specifically, when it is detected that the interaction between the time dimension constraint factor (such as the acceleration of approval of a certain node) and the space dimension constraint factor (such as the node's authority and responsibility exceeding the boundary) causes negative curvature, the curvature compensation algorithm will dynamically adjust the weight parameters in the fusion formula so that the spatiotemporal coupling features maintain local flatness in the manifold space. By compensating for single-mode negative curvature, the system can more accurately quantify the interactive effects of timing and spatial confidence (such as the amplified impact of timeouts in a certain link on the transfer of rights and responsibilities of subsequent nodes), avoiding the distortion of semantic shift coefficient calculation caused by spatial distortion; in addition, through the fusion space based on approximation to Euclideanness, the spatiotemporal weight distribution of process nodes is closer to the real business logic (for example, "cross-departmental collaboration on urgent matters" must meet both short-term efficiency and authority compliance), so that the final actual process transfer spatiotemporal coordination constraint factor can not only capture the dynamic rhythm of process advancement, but also map the implicit power gradient in the organizational structure.

[0049] In this example, an initial encoding vector is first passed based on each actual process sequence Corresponding actual process timing confidence constraint factor and the actual process space confidence constraint factor , to construct a constant curvature space representation and spherical coordinates approximation :

[0050]

[0051]

[0052] Then, the fused actual process is passed to the spatiotemporal synergy constraint factor As the fusion space metric benchmark, the weight parameter in the fusion formula is determined by calculating the following formula, for example and :

[0053]

[0054] Specifically, when When , the single-mode correlation in time and space dimensions can be expressed and The space-time coupling attraction tends to be flat, thus reflecting the Euclidean property in the fusion space. Specifically, the plane maintenance in the fusion space is achieved by compensating for the generation of single-mode negative curvature, thereby improving the space-time synergy constraint factor of the actual process transmission. fusion expression effect.

[0055] Furthermore, based on the actual process transmission spatiotemporal coordination constraint factor, the actual process sequence transmission aggregation analysis is performed on the time queue of the coding features of the key nodes of the actual process execution to obtain the actual process execution path semantic coding vector as the actual process execution path semantic coding feature. Specifically, in an embodiment of the present application, first, based on the actual process transmission spatiotemporal coordination constraint factor, the actual process transmission structure modulation is performed on each actual process sequence transmission initial coding vector to obtain the time series of the actual process sequence transmission structural modulation coding vector. Since there is significant heterogeneity in the spatiotemporal value of different nodes in the execution trajectory of the actual business process, for example, in the budget approval process of cross-departmental collaboration, the timeliness of key nodes (such as the review period of the financial department) and the authority compliance (such as the risk of over-level approval) are often intertwined, but the traditional system adopts equalization processing for all node information, resulting in the dilution of important signals by redundant operation data. In the technical solution of the present application, by introducing the spatiotemporal coordination constraint factor of the actual process transmission, the system can break through the limitations of static weight allocation, and dynamically adjust the intensity and direction of information transmission according to the differences in the importance of nodes in the spatiotemporal dimension (such as the timing confidence decay triggered by the timeout of a certain link, or the spatial weight anomaly caused by cross-level operations). In the process of structural modulation of message transmission, the system can adaptively strengthen the coding vectors of high-risk nodes (such as abnormally delayed approval links), while weakening the background noise of routine operations (such as standardized file archiving), so that the time series of the structural modulation coding vectors of the actual process sequence transmission generated in the end focuses on the key semantic features in the process execution (such as violations of regulations or efficiency bottlenecks). Through information purification under spatiotemporal coordination constraints, the accuracy and effectiveness of information transmission are enhanced.

[0056] Furthermore, calculate the position-wise sum of the time series of the structural modulation coding vectors passed by the actual process sequence to obtain the actual process execution path semantic coding vector as the actual process execution path semantic coding feature. Specifically, the execution path of the actual business process is used as time series data, and its coding vector has carried the dynamic features of spatio-temporal collaborative constraint modulation during the time series message passing process (such as abnormal fluctuations of key nodes and the power and responsibility relationships of cross-departmental collaborations). However, these features scattered in the time dimension are difficult to directly compare with the static semantic embedding features of the benchmark process effectively. Through the position-wise sum operation, the system can compress the dynamic trajectory information of the entire life cycle of the process execution (such as the spatio-temporal coupling features of each link from project establishment to payment in the budget approval process) into the actual process execution path semantic coding vector with global semantic representation ability to solve the problem of the mismatch between the time series data and the static rule matching dimension in traditional methods. In this process, the position-wise sum operation is not simply the accumulation of time dimension information, but on the basis of spatio-temporal collaborative constraint modulation, the distillation and sublimation of process semantics are realized through vector superposition in the feature space. Specifically, the structural modulation coding vector passed by the actual process sequence at each time step (such as the spatio-temporal weight feature of the "financial review" node) retains its position feature during the summation, so that the finally generated actual process execution path semantic coding vector not only reflects the overall execution trend of the process, but also can reveal the abnormal contribution degree of key nodes through the numerical distribution of each dimension of the vector.

[0057] In a specific example, based on the spatio-temporal collaborative constraint factor passed by the actual process, the following aggregation formula is used to perform the actual process sequence transfer aggregation analysis on the time queue of the coding features of the key nodes of the actual process execution to obtain the actual process execution path semantic coding vector; wherein, the aggregation formula is:

[0058]

[0059] Wherein, is the spatio-temporal collaborative constraint factor passed by the actual process, is the scale of the time series of the structural modulation coding vectors passed by the actual process sequence, is the actual process execution path semantic coding vector.

[0060] In particular, the S5 determines whether to generate a path deviation warning prompt based on the process execution semantic deviation between the actual process execution path semantic coding features and the benchmark process semantic embedding coding features. Specifically, in an embodiment of the present application, first, the process execution semantic deviation coefficient between the actual process execution path semantic coding vector and the benchmark process semantic embedding coding vector is calculated. Specifically, violations of actual business processes are often not manifested as explicit missing steps, but are hidden in deep semantic deviations such as time sequence logic dislocation and distortion of power and responsibility relationships (such as the seemingly complete approval chain in the procurement process actually has the authority to cross-level approval). Traditional monitoring methods based on rule matching or keyword retrieval are difficult to penetrate the process surface to capture such structural deviations at the semantic level. Through the calculation of geometric relationships in vector space (such as cosine similarity or Mahalanobis distance), the system can convert the semantic differences between institutional norms and actual execution into quantifiable deviation coefficients. The calculation of deviation coefficients is essentially a topological consistency assessment of the process execution path and institutional norms at the semantic level. By calculating the process execution semantic deviation coefficient between the semantic encoding vector of the actual process execution path and the semantic embedding encoding vector of the benchmark process, the semantic correlation analysis between the benchmark process and the actual process can be realized. In this process, the system can quantify the deviation between the actual process and the benchmark process in terms of temporal logic (such as inversion of approval order), authority compliance (such as cross-departmental unauthorized operations) and execution connotation (such as deviation in clause interpretation) through distance measurement in vector space. This relationship quantification based on geometric space provides data support for process optimization, thereby providing explainable and quantifiable decision support for dynamic and intelligent internal control management.

[0061] Furthermore, based on the comparison between the process execution semantic deviation coefficient and the preset threshold, it is determined whether to generate a path deviation warning prompt. In the technical solution of the present application, through the dynamic threshold comparison mechanism, the system can convert the abstract semantic deviation into a specific risk level signal, wherein the preset threshold is not a static value, but a dynamic decision boundary formed by combining business process attributes (such as capital scale, risk level), historical violation patterns (such as high-incidence links of unauthorized approval) and institutional tolerance (such as elastic space for special matters). In one example, when the process execution semantic deviation coefficient exceeds the threshold, the system triggers a real-time warning. In this way, it promotes the transformation and upgrading of the governance model of the internal control system of administrative institutions.

[0062] In summary, the AI-based internal control process optimization and supervision method for administrative institutions according to the embodiments of the present application is elucidated. By setting a benchmark process as a standard template, combining with the real-time data stream of the actual business system, and performing sequential message passing encoding on both the benchmark process and the actual process to achieve deep semantic alignment of the benchmark and actual paths, further quantifying the process execution semantic deviation coefficient between the benchmark process and the actual process after semantic alignment, and based on the comparison between the process execution semantic deviation coefficient and a preset threshold, determining whether to generate a path deviation warning prompt. In this way, the system can capture process deviations in real time and trigger warnings, enhancing the real-time and objectivity of supervision. At the same time, through the intelligent warning mechanism, it effectively prevents abnormal risks in process execution, providing administrative institutions with dynamic control capabilities that are both flexible and normative.

[0063] Furthermore, an AI-based internal control process optimization and supervision system for administrative institutions is also provided.

[0064] Figure 3 The block diagram of the AI-based internal control process optimization and supervision system for administrative institutions according to the embodiments of the present application. As Figure 3 shown, the AI-based internal control process optimization and supervision system 300 according to the embodiments of the present application includes: a benchmark setting module 310 for setting a benchmark process; a benchmark process encoding module 320 for performing semantic embedding encoding on each key node benchmark description in the benchmark process to obtain a time queue of benchmark process key node semantic embedding encoding features; an actual process encoding module 330 for obtaining the actual process execution path and performing structured encoding on each actual process execution key data in the actual process execution path to obtain a time queue of actual process execution key node encoding features; a sequential message passing encoding module 340 for performing sequential message passing encoding on the time queue of actual process execution key node encoding features and the time queue of benchmark process key node semantic embedding encoding features to obtain an actual process execution path semantic encoding feature and a benchmark process semantic embedding encoding feature; and a warning prompt module 350 for determining whether to generate a path deviation warning prompt based on the process execution semantic deviation degree between the actual process execution path semantic encoding feature and the benchmark process semantic embedding encoding feature.

[0065] As described above, the AI-based internal control process optimization and supervision system 300 for administrative institutions according to the embodiments of the present application can be implemented in various wireless terminals, such as a server with an AI-based internal control process optimization and supervision algorithm for administrative institutions. In a possible implementation manner, the AI-based internal control process optimization and supervision system 300 for administrative institutions according to the embodiments of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the AI-based internal control process optimization and supervision system 300 can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the AI-based internal control process optimization and supervision system 300 can also be one of the numerous hardware modules of the wireless terminal.

[0066] Alternatively, in another example, the AI-based internal control process optimization and supervision system 300 and the wireless terminal can also be separate devices, and the AI-based internal control process optimization and supervision system 300 can be connected to the wireless terminal through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0067] The embodiments of the present disclosure have been described above. The above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technologies in the market, or to enable other ordinary skill in the art in the technical field to understand the disclosed embodiments.

Claims

1. An AI-based internal control process optimization and supervision method for administrative institutions, characterized by: include: Setting the benchmark process; Perform semantic embedding coding on the benchmark descriptions of each key node in the benchmark process to obtain a time queue of semantic embedding coding features of the key nodes of the benchmark process; Acquire the actual process execution path, and perform structured encoding on each actual process execution key data in the actual process execution path to obtain a time queue of encoding features of the actual process execution key nodes; Performing temporal message transmission encoding on the time queue of encoding features of key nodes of actual process execution and the time queue of semantic embedding encoding features of key nodes of benchmark process to obtain semantic encoding features of actual process execution path and semantic embedding encoding features of benchmark process, including: respectively calculating process node spatiotemporal collaborative confidence constraint factors of the time queue of encoding features of key nodes of actual process execution and the time queue of semantic embedding encoding features of key nodes of benchmark process to construct a process semantic transmission modulation aggregation mechanism, and performing sequence transmission aggregation on the time queue of encoding features of key nodes of actual process execution and the time queue of semantic embedding encoding features of key nodes of benchmark process based on the process semantic transmission modulation aggregation mechanism to obtain semantic encoding features of actual process execution path and semantic embedding encoding features of benchmark process; Based on the process execution semantic deviation between the actual process execution path semantic coding features and the benchmark process semantic embedding coding features, it is determined whether to generate a path deviation warning prompt.

2. According to the AI-based internal control process optimization and supervision method for administrative institutions according to claim 1, it is characterized in that: The semantic embedding encoding is performed on the benchmark descriptions of each key node in the benchmark process to obtain a time queue of semantic embedding encoding features of the key nodes of the benchmark process, including: The benchmark descriptions of each key node in the benchmark process are input into the benchmark process node semantic encoder based on the Bert model to obtain the time queue of the benchmark process key node semantic embedding encoding vectors as the time queue of the benchmark process key node semantic embedding encoding features.

3. According to claim 2, the AI-based internal control process optimization and supervision method for administrative institutions is characterized in that: The actual process execution path is obtained, and each actual process execution key data in the actual process execution path is structuredly encoded to obtain a time queue of encoding features of the actual process execution key nodes, including: receiving actual process execution key data from an actual business system to obtain an actual process execution path composed of a time queue of the actual process execution key data; The actual process embedding matrix is ​​used to perform structured encoding on each actual process execution key data in the actual process execution path to obtain a time queue of the actual process execution key node encoding vector as the time queue of the actual process execution key node encoding feature.

4. According to claim 1, the method for optimizing and supervising the internal control process of administrative institutions based on AI is characterized in that: The time queue of the encoding features of the key nodes of the actual process execution and the time queue of the semantic embedding encoding features of the key nodes of the benchmark process are encoded by time-series message passing to obtain the semantic encoding features of the actual process execution path and the semantic embedding encoding features of the benchmark process, including: Calculate the spatiotemporal coordination confidence constraint factors of the process nodes of the time queue of the encoding features of the key nodes of the actual process execution to construct the spatiotemporal coordination constraint factors of the actual process delivery; Calculate the spatiotemporal coordination confidence constraint factors of the process nodes of the time queue of the semantic embedding encoding features of the key nodes of the benchmark process to construct the spatiotemporal coordination constraint factors of the benchmark process transfer; Based on the spatiotemporal coordination constraint factor of the actual process transmission, the actual process sequence transmission aggregation analysis is performed on the time queue of the encoding features of the key nodes of the actual process execution to obtain the semantic encoding vector of the actual process execution path as the semantic encoding feature of the actual process execution path; Based on the benchmark process transfer spatiotemporal coordination constraint factor, the actual process sequence transfer aggregation analysis is performed on the time queue of the semantic embedding coding features of the key nodes of the benchmark process to obtain the benchmark process semantic embedding coding vector as the benchmark process semantic embedding coding feature.

5. According to claim 4, the AI-based internal control process optimization and supervision method for administrative institutions is characterized in that: Calculate the spatiotemporal coordination confidence constraint factor of the process node of the time queue of the encoding features of the key nodes of the actual process execution to construct the spatiotemporal coordination constraint factor of the actual process delivery, including: Input the time queue of the encoding vectors of the key nodes of the actual process execution into the process node sequence encoder based on the recurrent neural network to obtain the time sequence of the initial encoding vector transmitted by the actual process sequence; Calculate the actual process timing confidence constraint factor of each actual process sequence transmitting the initial coding vector in the time series of the actual process sequence transmitting the initial coding vector; Calculate the actual process space confidence constraint factor of each actual process sequence transmitting the initial coding vector in the time series of the actual process sequence transmitting the initial coding vector; Based on the actual process space confidence constraint factor and the actual process time sequence confidence constraint factor of each actual process sequence transmitting the initial coding vector, the actual process transmission space-time coordination constraint factor of each actual process sequence transmitting the initial coding vector is constructed.

6. The AI-based internal control process optimization and supervision method for administrative institutions according to claim 5 is characterized in that: Based on the actual process space confidence constraint factor and the actual process time sequence confidence constraint factor of each actual process sequence transmitting the initial coding vector, the actual process transmission time and space coordination constraint factor of each actual process sequence transmitting the initial coding vector is constructed, including: The actual process space confidence constraint factor and the actual process time sequence confidence constraint factor are integrated to obtain the initial actual process transfer time and space coordination constraint factor; The initial actual process transfer space-time coordination constraint factor is optimized by space-time curvature compensation to obtain the actual process transfer space-time coordination constraint factor.

7. The AI-based internal control process optimization and supervision method for administrative institutions according to claim 6 is characterized in that: Based on the spatiotemporal coordination constraint factors of the actual process transmission, the actual process sequence transmission aggregation analysis is performed on the time queue of the encoding features of the key nodes of the actual process execution to obtain the semantic encoding vector of the actual process execution path as the semantic encoding feature of the actual process execution path, including: Based on the spatiotemporal synergy constraint factor of actual process transmission, actual process transmission structure modulation is performed on each actual process sequence transmission initial coding vector to obtain a time series of actual process sequence transmission structural modulation coding vectors; The position-wise sum of the time series of the structural modulation coding vector transferred by the actual process sequence is calculated to obtain the actual process execution path semantic coding vector as the actual process execution path semantic coding feature.

8. The AI-based internal control process optimization and supervision method for administrative institutions according to claim 7 is characterized in that: Based on the process execution semantic deviation between the actual process execution path semantic coding features and the benchmark process semantic embedding coding features, determine whether to generate a path deviation warning prompt, including: Calculate the process execution semantics offset coefficient between the actual process execution path semantic encoding vector and the benchmark process semantic embedding encoding vector; Based on the comparison between the process execution semantic deviation coefficient and the preset threshold, it is determined whether to generate a path deviation warning prompt.

9. The AI-based internal control process optimization and supervision system for administrative institutions is characterized by: include: Benchmark setting module, used to set the benchmark process; A benchmark process encoding module, used for semantically embedding the benchmark descriptions of each key node in the benchmark process to obtain a time queue of semantically embedded encoding features of the key nodes of the benchmark process; The actual process encoding module is used to obtain the actual process execution path and perform structured encoding on each actual process execution key data in the actual process execution path to obtain a time queue of encoding features of the actual process execution key nodes; A time series message passing encoding module is used to perform time series message passing encoding on the time queue of the encoding features of the key nodes of the actual process execution and the time queue of the semantic embedding encoding features of the key nodes of the benchmark process to obtain the semantic encoding features of the actual process execution path and the semantic embedding encoding features of the benchmark process; The early warning prompt module is used to determine whether to generate a path deviation early warning prompt based on the process execution semantic deviation between the actual process execution path semantic coding features and the benchmark process semantic embedding coding features.

Images