A Method and System for Internal Control Management Decision-Making Based on Multimodal Data Analysis

Abstract

This invention relates to the field of data decision analysis, and more particularly to a method and system for unit internal control management decision-making based on multimodal data analysis. The method includes the following steps: collecting heterogeneous raw data streams from multiple heterogeneous data sources; performing spatiotemporal entropy change differential analysis and multi-dimensional entropy change characteristic mining; performing multi-level semantic analysis on the heterogeneous raw data streams and performing layer-by-layer abnormal pattern semantic deconstruction to construct a precise semantic feature profile for each abnormal pattern; performing multi-objective coordination request analysis and dynamic objective priority evaluation on the cognitive entropy change benchmark feature map to obtain the coordination request priority; calculating the maximum decision capacity of the cognitive entropy change benchmark feature map, and calculating the dynamic optimal decision boundary based on the coordination request priority and the precise semantic feature profile, thereby generating a dynamic trade-off constraint range. This invention smooths decision objective conflicts and improves the efficiency of user decision-making needs execution through multi-objective collaborative decision analysis.

Description

Technical Field

[0001] This invention relates to the field of data decision analysis, and in particular to a method and system for internal control management decision-making based on multimodal data analysis. Background Technology

[0002] With the continuous improvement of information technology infrastructure and digital governance, organizations are increasingly reliant on data in business management, risk control, and decision support. Especially in internal control management, the comprehensiveness, accuracy, and intelligent processing capabilities of data have become key indicators for measuring the efficiency and reliability of internal control systems. At the same time, core management tasks such as financial auditing, risk prevention, and performance evaluation also place higher demands on the intelligence and refinement of internal control mechanisms. Traditional internal control management methods mainly rely on periodic manual audits, tabular report analysis, and static indicator evaluation. These methods have significant shortcomings in data collection, risk identification, and anomaly warning, making it difficult to respond promptly to complex and ever-changing business scenarios and unable to effectively integrate various heterogeneous data sources to support comprehensive decision-making. Furthermore, faced with massive amounts of data and highly fragmented information, traditional internal control methods often lack intelligent linkage mechanisms and dynamic analysis capabilities, leading to lagging risk control, severe information silos, and ultimately affecting the organization's operational efficiency and management level.

[0003] With the rapid development of emerging technologies such as big data, artificial intelligence, and the Internet of Things, various organizations have gradually accumulated a large amount of structured data (such as financial data, approval records, and personnel files) and unstructured data (such as meeting minutes, voice recordings, email content, and surveillance videos) in their daily management processes. This data exhibits multimodal characteristics, including diverse sources, complex formats, and rapid updates, providing a new opportunity for in-depth internal control management analysis and decision support. However, current mainstream internal control systems often lack a unified data fusion mechanism, failing to effectively integrate and deeply mine multimodal data. This results in hidden risks being difficult to identify in a timely manner and cannot support higher-dimensional, more forward-looking decision-making. Therefore, a more intelligent internal control management decision-making method is urgently needed. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a decision-making method and system for unit internal control management based on multimodal data analysis, thereby resolving at least one of the aforementioned technical problems.

[0005] To achieve the above objectives, this invention provides a decision-making method for internal control management based on multimodal data analysis, comprising the following steps:

[0006] Step S1: Collect heterogeneous raw data streams from the unit's internal control management platform; perform spatiotemporal entropy change differential analysis and multi-dimensional entropy change characteristic mining to generate a cognitive entropy change benchmark feature map;

[0007] Step S2: Perform multi-level semantic parsing on the heterogeneous raw data stream, and perform layer-by-layer abnormal pattern semantic deconstruction to construct a precise semantic feature profile for each abnormal pattern;

[0008] Step S3: Obtain the preset decision target instruction, perform multi-target coordination request analysis and dynamic target priority evaluation on the cognitive entropy change benchmark feature map, and obtain the coordination request priority;

[0009] Step S4: Calculate the maximum decision capacity of the cognitive entropy change baseline feature map, and calculate the dynamic optimal decision boundary based on the coordination request priority and accurate semantic feature profile, thereby generating a dynamic trade-off constraint range.

[0010] Step S5: Solve the multi-objective dynamic constraint response based on the dynamic trade-off constraint range, and perform real-time decision processing to obtain the real-time execution effect;

[0011] Step S6: Identify the deviation of decision execution response based on the real-time execution effect, and perform in-depth optimization of strategy iteration to build an intelligent decision-making body for internal control management.

[0012] This specification provides a multimodal data analysis-based internal control management decision-making system for executing the multimodal data analysis-based internal control management decision-making method described above, including:

[0013] The data stream entropy change analysis module is used to collect heterogeneous raw data streams from the unit's internal control management platform; perform spatiotemporal entropy change differential analysis and multi-dimensional entropy change characteristic mining to generate a cognitive entropy change benchmark feature map;

[0014] The semantic parsing module is used to perform multi-level semantic parsing on the heterogeneous raw data stream, and to perform layer-by-layer abnormal pattern semantic deconstruction to construct a precise semantic feature profile for each abnormal pattern.

[0015] The multi-objective coordination module is used to obtain preset decision-making objective instructions, perform multi-objective coordination request analysis and dynamic objective priority evaluation on the cognitive entropy change benchmark feature map, and obtain the coordination request priority.

[0016] The decision boundary module is used to calculate the maximum decision capacity of the cognitive entropy change benchmark feature map, and to calculate the dynamic optimal decision boundary based on the coordination request priority and accurate semantic feature profile, thereby generating a dynamic trade-off constraint range.

[0017] The constraint solving module is used to solve multi-objective dynamic constraint responses based on the dynamic trade-off constraint range, and to perform real-time decision processing to obtain real-time execution results.

[0018] The internal control management decision module is used to identify deviations in decision execution response based on real-time execution results, and to perform in-depth optimization of strategy iteration to build an intelligent internal control management decision-making body.

[0019] The beneficial effects of this invention are as follows: By collecting heterogeneous raw data streams from multiple heterogeneous data sources and performing entropy change differential analysis in the spatiotemporal dimensions and mining multi-dimensional entropy change characteristics, high-precision dynamic cognitive modeling of the system's operating state can be achieved. Entropy change, as a sensitive indicator of system complexity and uncertainty, can reveal potential instability or trend shifts in data changes, thereby helping the system identify potential abnormal signals in the early stages. The generated cognitive entropy change baseline feature map possesses high distinguishability and stability, becoming a basic reference model for subsequent anomaly detection and decision optimization, effectively improving overall perception sensitivity and response foresight. By performing multi-level semantic analysis on heterogeneous raw data streams and deconstructing the anomaly patterns contained within them layer by layer, the unique behavioral characteristics, evolutionary laws, and system impact paths of each type of anomaly can be deeply extracted, constructing a highly accurate semantic feature profile. This semantic deconstruction process breaks through the traditional rule-based or threshold-based anomaly identification methods, enabling the system to identify and classify "novel, latent, and gradual" anomalies. Precise semantic feature profiling not only improves the accuracy of anomaly classification but also provides a rich and structured information foundation for subsequent decision-making trade-offs and prioritization. By receiving pre-set decision-making target instructions and combining entropy-change feature maps to conduct multi-target coordination request analysis and dynamic priority evaluation, the system can effectively solve the problems of conflict, resource competition, and execution order in multi-target decision-making processes. Through the generation of dynamic coordination request priorities, the system can adapt to the urgency, benefit value, and execution cost of targets in different scenarios, achieving a more flexible and intelligent target management mechanism. This process significantly improves the system's scheduling capability and execution stability under conditions of limited resources and frequent anomalies. By calculating the maximum decision capacity of the entropy-change feature map and combining it with the generated coordination request priorities and anomaly semantic profiling, a dynamic and accurate estimation of the decision space boundary that the system can bear in the current state can be achieved. The generated dynamic optimal decision boundary not only reflects the system's true perception of current risks and resources but also significantly enhances the rationality and risk resistance of the strategy generation process through dynamic trade-off constraints, avoiding decision collapse caused by overload or erroneous scheduling and ensuring the system's strategy security in complex and ever-changing environments. By solving for multi-objective dynamic constraint responses within a dynamic trade-off constraint range and triggering real-time decision processing, it can be ensured that the response strategy, when multiple objectives exist, both conforms to system limitations and maximizes objective benefits. The core of this step lies in achieving "execution-while-optimization" of decisions, ensuring that the decision-making process has the ability to adapt instantly to environmental conditions, abnormal fluctuations, and priority changes. Its direct effect is to improve system execution efficiency, stability, and the accuracy of strategy implementation, making it particularly suitable for complex scenarios with rapid dynamic changes and high decision-making frequency, such as energy scheduling and real-time traffic control.By systematically identifying deviations in real-time execution performance and embedding them as feedback signals into the strategy optimization process, a continuously evolving, self-learning intelligent agent for internal control management decision-making can be constructed. This agent possesses a closed-loop learning capability of "feedback-correction-re-optimization," continuously correcting historical deviations, absorbing execution experience, and enhancing the adaptability and robustness of future strategies. Ultimately, this results in an intelligent decision-making engine that can continuously adapt to the dynamic changes of complex systems, possessing highly adaptive and self-recovering capabilities, fundamentally improving the system's stability, decision-making efficiency, and sustainable optimization capabilities. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the steps of a unit internal control management decision-making method based on multimodal data analysis according to the present invention.

[0021] Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1.

[0022] Figure 3 This is a detailed flowchart illustrating the implementation steps of step S2;

[0023] Figure 4 This is a flowchart illustrating the detailed implementation steps of step S3. Detailed Implementation

[0024] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0025] This application provides a method and system for internal control management decision-making based on multimodal data analysis. The executing entities of the method and system include, but are not limited to, the following: mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc., which can be considered general computing nodes of this application. The data processing platform includes, but is not limited to, at least one of the following: an audio-visual management system, an information management system, and a cloud-based data management system.

[0026] Please see Figures 1 to 4 This invention provides a decision-making method for internal control management based on multimodal data analysis, which includes the following steps:

[0027] Step S1: Collect heterogeneous raw data streams from the unit's internal control management platform; perform spatiotemporal entropy change differential analysis and multi-dimensional entropy change characteristic mining to generate a cognitive entropy change benchmark feature map;

[0028] Step S2: Perform multi-level semantic parsing on the heterogeneous raw data stream, and perform layer-by-layer abnormal pattern semantic deconstruction to construct a precise semantic feature profile for each abnormal pattern;

[0029] Step S3: Obtain the preset decision target instruction, perform multi-target coordination request analysis and dynamic target priority evaluation on the cognitive entropy change benchmark feature map, and obtain the coordination request priority;

[0030] Step S4: Calculate the maximum decision capacity of the cognitive entropy change baseline feature map, and calculate the dynamic optimal decision boundary based on the coordination request priority and accurate semantic feature profile, thereby generating a dynamic trade-off constraint range.

[0031] Step S5: Solve the multi-objective dynamic constraint response based on the dynamic trade-off constraint range, and perform real-time decision processing to obtain the real-time execution effect;

[0032] Step S6: Identify the deviation of decision execution response based on the real-time execution effect, and perform in-depth optimization of strategy iteration to build an intelligent decision-making body for internal control management.

[0033] In the embodiments of the present invention, see Figure 1 This is a flowchart illustrating the steps of a unit internal control management decision-making method based on multimodal data analysis according to the present invention. In this example, the steps of the method include:

[0034] Step S1: Collect heterogeneous raw data streams from the unit's internal control management platform; perform spatiotemporal entropy change differential analysis and multi-dimensional entropy change characteristic mining to generate a cognitive entropy change benchmark feature map;

[0035] In this embodiment, after obtaining authorization from relevant departments and users, the core of the first step in the internal control management decision-making scenario of the unit is to systematically collect heterogeneous raw data streams from multiple business modules, conduct in-depth information entropy analysis on these data streams, combine the dynamic evolution characteristics of the spatiotemporal dimension, and finally extract multi-dimensional entropy change features to construct a set of cognitive-level benchmark feature maps, providing a basic reference for subsequent intelligent monitoring, resource scheduling and decision optimization.

[0036] During the initial data acquisition phase, data integration was required across the organization's six core business systems: the budget management system, revenue management platform, expenditure settlement system, procurement management platform, contract execution system, and materials and assets management system. Each system's data interface exhibited structural heterogeneity; some used traditional Oracle databases, others provided web services or REST APIs, and still others could only be accessed through log monitoring or data snapshots. Data extraction and integration tools (such as Apache NiFi or Talend) were employed, combined with a unified data adapter, to connect to the data sources and establish a field mapping model based on data elements. For example, the "budget item number" in the budget system needed to be cross-referenced with the "expenditure item code" in the expenditure system to ensure consistency in subsequent analysis. The data acquisition granularity was set to "hourly" to guarantee temporal continuity and analytical accuracy.

[0037] After data collection, the process proceeds to the information entropy density calculation and entropy change differential analysis stage. For each type of heterogeneous raw data stream, a data frame is first constructed with time as the main axis. For example, budget data is aggregated by "month," income and expenditure data by "day," contract and procurement data is discretized by "approval node," and material data is formed into an event sequence according to "inventory fluctuations." Next, the entropy value is calculated for the state distribution of each type of data within different time windows using Shannon's information entropy formula H = -∑p(x)logp(x), where...

[0038] p(x) represents the probability distribution of a certain indicator (such as budget execution rate or procurement plan completion rate) under a specific state. To perceive the spatiotemporal dynamic evolution trend of entropy, the entropy change differential method is further employed to calculate the gradient rate of change of information entropy between adjacent time periods or adjacent organizational units, thereby constructing a set of data entropy evolution trajectories with temporal characteristics. In the experiment, a sliding time window of 3 days was set, and a first-order difference algorithm was used to process the continuous entropy sequence. It was found that the materials system exhibits obvious entropy change peaks at the beginning and end of the month, reflecting periodic allocation fluctuations.

[0039] Next, we move into the data topology and multi-dimensional entropy change characteristic mining stage. By constructing a relational topology structure between various types of business data, we establish a causal path and collaborative relationship network between data flows. For example, budget data influences procurement plans, and procurement completion, in turn, affects expenditure execution; contract performance is simultaneously affected by both budget freezes and material delivery status. Based on Graph Neural Networks (GNNs), we perform node embedding learning to extract the dynamic coupling characteristics between various data sources and superimpose entropy change differential features to form a multimodal embedding feature matrix. Finally, we use dimensionality reduction methods such as t-SNE or UMAP to map the high-dimensional features to a two-dimensional space, creating a cognitive entropy change benchmark feature map. This map uses a "data channel-time period" coordinate system, where each point corresponds to the entropy change state of a data subsystem within a specific time window. Color intensity represents the entropy value, and cluster density reflects the system's complexity and abnormal evolution trends.

[0040] Step S2: Perform multi-level semantic parsing on the heterogeneous raw data stream, and perform layer-by-layer abnormal pattern semantic deconstruction to construct a precise semantic feature profile for each abnormal pattern;

[0041] In this embodiment, after constructing the cognitive entropy change baseline map during the unit's multimodal data analysis process, the next key step is to perform multi-level semantic parsing on the heterogeneous raw data streams. This parsing, combined with historical and real-time behavioral trajectories, allows for in-depth analysis of potential abnormal patterns, thereby providing support for risk identification and response strategies in internal control management. This step aims to construct an abnormal pattern recognition and classification system with semantic interpretation capabilities. By reconstructing and analyzing the behavioral trajectories and data semantics of each abnormal form, a fine-grained semantic feature profile is generated.

[0042] First, the starting point for multi-level semantic parsing lies in defining the semantic hierarchy of the data. In this scenario, the raw data includes six core business categories: budget, revenue, expenditure, procurement, contracts, and materials management. Each category contains structured numerical values ​​(such as amount, percentage, and frequency) and weakly structured text (such as contract content, reasons for procurement applications, and budget explanations). A unified semantic framework needs to be established based on the hierarchical structure of "semantic entity—attribute—behavior." For example, in contract data, "abnormal contract amount," "fluctuation in performance cycle," and "abnormal payment node" are identified as first-level semantic entities. Second-level attributes include "amount deviation rate" and "node time offset," while third-level behaviors can be abstracted into behavioral labels such as "contract execution violates the original budget plan." This stage mainly uses natural language processing tools such as BERT or RoBERTa pre-trained models to extract contextual semantic vectors from weakly structured fields. For structured fields, a label mapping mechanism is used to uniformly abstract them into standard business behavior language (for example, converting "budget execution rate <40%" into the label "severe underperformance").

[0043] Next, the system proceeds to the layer-by-layer semantic deconstruction of abnormal patterns. It needs to identify potential abnormal patterns in multi-source data, such as "contract signing and budget freeze times are not synchronized," "purchase amount exceeds approved limit," and "sudden changes in material requisition frequency." To achieve high accuracy, an anomaly detection algorithm based on Graph Attention Network (GAT) is introduced. This model models the processes and approval relationships between business operations as a heterogeneous graph structure, where nodes represent various business entities (such as purchase orders, budgets, invoices, contracts, etc.), and edges represent the temporal logic or cash flow relationships between entities. In this graph model, abnormal subgraph structures are identified by weighted aggregation of node attribute features and their neighbor relationships. For example, when the purchase amount in an approval path is continuously higher than 30% of similar contracts, and the approval time is significantly lower than the average, this subpath will be labeled as "high-risk fast procurement."

[0044] During the deconstruction process, the underlying data semantic chain is further reconstructed for each identified anomaly pattern. For example, an abnormal procurement process may involve multiple data anomaly nodes such as delayed budget allocation, inconsistent contract signing, and delayed inventory records. Through causal chain modeling and semantic backtracking, these data entities are arranged chronologically and semantically labeled to construct an "anomaly pattern semantic chain." In the experiment, based on the collected samples exported from the 2024 financial system of a prefecture-level city, approximately 812 anomaly patterns were identified, with an average of 3.2 semantic nodes per pattern, involving time spans ranging from several hours to a week. Using similarity measurement methods (such as cosine similarity or KL divergence) to perform cluster analysis on the anomaly semantic chains, typical anomaly categories such as "rapid jumps in budget-procurement" and "fragmented fluctuations in contract amount" can be formed.

[0045] Finally, a precise semantic feature profile of the anomaly patterns is constructed. Based on the above analysis, the following semantic feature vectors are extracted for each anomaly type: anomaly triggering conditions (such as amount, time threshold), participating data stream category, key behavioral nodes, time span, anomaly persistence, number of collaborative anomalies, and historical frequency of occurrence. These features are standardized and stored in the semantic profile template to form a reusable and comparable anomaly type knowledge base. Based on this, organizations can achieve rapid identification of similar anomalies, automatic hierarchical handling, and early warning for internal control audits.

[0046] Step S3: Obtain the preset decision target instruction, perform multi-target coordination request analysis and dynamic target priority evaluation on the cognitive entropy change benchmark feature map, and obtain the coordination request priority;

[0047] In this embodiment, during the internal control management process of an enterprise, different management modules (such as budget control, procurement approval, contract performance, expenditure monitoring, etc.) often have multiple decision-making objectives operating in parallel. These objectives may overlap in time, compete for resources, and even conflict in results. Therefore, after completing the construction of the cognitive entropy change baseline feature map and the anomaly semantic profile, the system needs to introduce a multi-objective coordination mechanism to dynamically respond to different objective instructions issued by the internal control management department, and accordingly evaluate the coordination priority of objective requests, thereby ensuring the rationality and efficiency of the entire data analysis decision-making chain. This step revolves around the core of "multi-objective regulation," realizing real-time dynamic priority evaluation based on the data entropy change map.

[0048] The system acquires pre-defined decision-making objectives and instructions. These instructions originate from the upper-level strategy deployment of the internal control management platform, including static objectives (such as improving quarterly budget execution efficiency to over 90%) and dynamic objectives (such as optimizing emergency expenditure approval paths and increasing the priority of procurement early warning responses). Each objective instruction is standardized into a multi-dimensional objective metadata structure, including objective category, associated data source, expected output result, impact period, weighting factors, and conflict constraints. For example, an instruction might include: "Control the over-budget procurement rate of materials to no more than 5% within this quarter (high priority), while ensuring that the timely contract fulfillment rate is greater than 95% (medium priority)."

[0049] Based on a cognitive entropy change baseline feature map, the system analyzes the correlation strength between each target request and different entropy change regions in the map. This process is handled through semantic mapping and indicator clustering. Specifically, the system uses indicator items in the target (such as "budget execution rate", "procurement frequency", and "contract performance deviation rate") as query vectors, and uses node features in the map (including entropy change trajectory, temporal gradient, and semantic entities) to calculate vector similarity, evaluating the coupling degree between the target and the current system state. In experiments, based on a simulated dataset from a provincial government affairs platform, after modeling the semantic items of the indicators using the TF-IDF weighted word vector method, the average matching accuracy with entropy change nodes reached 91.7%.

[0050] The system then enters the multi-objective coordination request analysis phase. Since multiple target instructions may map to the same data channel or adjacent entropy-change characteristic paths, the system needs to perform conflict identification and coordinated scheduling. During this process, a multi-objective optimization framework (such as the MOEA algorithm based on Pareto optimal solutions, Multi-objective Evolutionary Algorithm) is used to model and analyze the coupling, conflict, and resource consumption between multiple targets, generating an inter-target conflict matrix. Matrix elements represent the conflict intensity of two targets in a certain resource domain (such as approval process time, data computation occupation, signal channels, etc.). By performing spectral clustering on the conflict matrix, high-conflict target pairs requiring coordination can be quickly identified.

[0051] The system performs dynamic target priority assessment. The assessment methodology employs a joint modeling mechanism combining AHP (Analytic Hierarchy Process) and weighted regression. First, each target is scored using initial weights manually configured by managers. Then, system logs and historical data are used to perform a second round of weighted adjustments based on factors such as target achievement, execution impact, and urgency level, thereby generating a final priority sequence for coordination requests. For example, if "rapid response to procurement approvals" is found to occur frequently in historical anomalies and has a significant impact on budget and contract execution, then this target will have a significantly higher priority in the current cycle.

[0052] The final output of the coordination request priority result will be fed back to the strategy scheduling engine as the basic input for subsequent process control, resource allocation, and anomaly priority response, providing real-time goal-oriented optimization strategies for the entire organization's internal control management. Through this step, not only is the structured management of internal control objectives achieved, but the responsiveness of internal control strategies and the efficiency of resource coordination are also improved. In real-world scenarios, it has been verified that multi-objective response speed can be improved by approximately 36%, while significantly reducing the frequency of resource allocation conflicts.

[0053] Step S4: Calculate the maximum decision capacity of the cognitive entropy change baseline feature map, and calculate the dynamic optimal decision boundary based on the coordination request priority and accurate semantic feature profile, thereby generating a dynamic trade-off constraint range.

[0054] In this embodiment, within the organization's internal control management decision-making system based on multimodal data, as multiple business objectives are coordinated and scheduled, and semantic anomaly profiles are constructed, the system needs to quantitatively evaluate its overall decision-making capabilities. Based on this, a dynamic trade-off mechanism is built to achieve boundary control and resource allocation optimization among various objective instructions. The core objective of this step is to dynamically model the resource scheduling boundaries of the internal control system during actual operation by calculating the maximum decision-making capacity that the current system can bear, and combining the composite characteristics of coordination request priority and anomaly semantics. This generates a dynamic trade-off constraint range with time elasticity, resource matching, and semantic interpretability.

[0055] The first stage involves calculating the maximum decision-making capacity. This step, based on the previously constructed cognitive entropy change baseline feature map, identifies the upper limit of the processing capacity of each key data channel, entropy change node, and behavioral path in the system. Specifically, each business entropy change trajectory in the map (e.g., budget fluctuation frequency, expenditure variability, procurement response time) is treated as a "data decision-making unit," and parameters such as its average processing response time, data bandwidth consumption, and maximum load pressure within a historical time window are calculated to form a "unit decision-making energy index." Taking a municipal government platform as an example, statistical analysis of historical data in the budget module reveals that when the number of budget approval anomalies exceeds 1.8 times the daily average processing capacity, the system's anomaly detection accuracy drops to 78%, and system response latency increases by 22%. Combining this data, the system's current "decision pressure limit," i.e., the maximum decision-making capacity, is calculated, typically measured by the number of high-priority tasks that can be completed per unit time and their average execution quality.

[0056] The second stage is the calculation of the dynamic optimal decision boundary. This calculation integrates the coordination request priority results output from the previous step with the accurate semantic feature profile. Its core idea is that different priorities of objectives consume different amounts of system resources and processing power in different semantic anomaly scenarios. Therefore, it is necessary to construct a resource "consumption function" for each objective request based on the semantic profile. For example, the anomaly profile of "procurement contract delay risk" contains 3 types of behavioral paths and 6 data entity nodes, and the computational resources required for its processing account for more than 20% higher than those required for the "budget adjustment deviation" anomaly profile. Therefore, when modeling the multi-objective processing boundary, a multivariable function F(x) = w1·R1 + w2·R2 + ... + w n ·R n , where R n w represents the semantic complexity-weighted resource requirement for each abnormal request. n This represents the priority weight. The system obtains the optimal dynamic scheduling boundary among a set of objectives by solving for the optimal boundary point of this function under the maximum decision capacity.

[0057] The third stage involves generating the dynamic trade-off constraint range. Based on the calculation results of the dynamic optimal decision boundary mentioned above, the system defines the boundary conditions and resource switching thresholds that can be accommodated simultaneously between different objectives, forming a "dynamic trade-off constraint matrix." Each row in the matrix represents the trade-off range of a high-priority objective in the current state (i.e., maximum resource scheduling interval, minimum response latency requirement, acceptable objective drift, etc.), and each column represents the dependency response relationship of other objectives under its influence. For example, for the priority control objective of "increased risk of delayed revenue receipt," the system can allow it to occupy a maximum of 30% of bandwidth resources under the current data resource scheduling model, and limit its delay in the contract performance logic channel to no more than 2 hours; if this boundary is breached, an emergency rescheduling mechanism needs to be triggered.

[0058] The final output is a dynamic trade-off constraint range map covering various target instructions. This map not only reflects the system's resource scheduling boundaries and behavioral processing capabilities but can also be directly embedded into the internal control strategy engine for dynamic task allocation, automatic priority adjustment, and intelligent risk avoidance. In practical applications, by introducing this map, a municipal unit reduced the resource conflict rate in contract approval and procurement scheduling by 41% and decreased the overall average approval latency by 18%, significantly improving the unit's management flexibility and response efficiency under multi-task and multi-resource conflicts. This step provides a solid boundary foundation for subsequent intelligent strategy recommendation and risk response simulation.

[0059] Step S5: Solve the multi-objective dynamic constraint response based on the dynamic trade-off constraint range, and perform real-time decision processing to obtain the real-time execution effect;

[0060] In this embodiment, the system extracts the schedulable boundary information for the current moment from the "dynamic trade-off constraint range map" output in the previous stage. Each dynamic trade-off boundary reflects the range of resources that can be scheduled for a specific objective under a specific business state, the allowable response latency, and the priority dependency relationship with other objectives. For example, in a unit's budget scheduling system, the currently acceptable "budget execution excess fluctuation" is ±3.5%, the procurement response delay must not exceed 48 hours, and the concurrent processing of contract approvals is limited to within 3 paths. The system transforms these boundary conditions into constraint function input values ​​for subsequent solver modeling.

[0061] Secondly, the system enters the multi-objective dynamic constraint response solution stage. A multi-objective optimization mathematical model is constructed, treating each business objective as an objective function (such as minimizing approval delay, maximizing contract fulfillment efficiency, and controlling procurement overflow rate), with resource boundaries, delay constraints, and dependencies as input inequality constraints. To achieve efficient solution, the system can employ evolutionary algorithms (such as the NSGA-II-based multi-objective non-dominated sorting algorithm) or convex optimization techniques. Taking NSGA-II as an example, it uses a population evolution strategy to select, crossover, and mutate candidate solutions, generating a set of non-dominated solutions that satisfy the constraints, and then adjusts the objective bias based on objective priority and the current semantic profile importance. Experiments show that in scenarios with dense internal resource conflicts, this algorithm can achieve over 90% objective coverage and output an executable strategy within 100ms.

[0062] Subsequently, the system converts the scheduling parameters in the optimal solution into real-time decision instructions for invoking real-time business processes. This includes the following types of operations:

[0063] Prioritize scheduling of tasks initiated by the approval system along specific paths;

[0064] Escalate the response level for abnormal events (such as abnormally skipping the approval level for a purchase order);

[0065] Adjust the tolerance range of budget or contract data to maintain system stability;

[0066] Dynamically switch data collection priorities to ensure that high-value information is processed first.

[0067] When a government agency discovers that its procurement budget exceeds the threshold and that the contract performance time has deviated, the system outputs an immediate processing instruction set based on the constraint solution, which includes "freezing procurement instructions + scheduling contract early warning logic + extending the contract tolerance time window by 4 hours," and immediately pushes it to the business middle platform.

[0068] The system collects feedback data in real time from each business path after execution, forming a preliminary set of real-time execution performance evaluation indicators. This evaluation set mainly includes key indicators such as response timeliness (e.g., approval acceleration rate), resource utilization, deviation from target achievement, and changes in task conflict frequency. By comparing these indicators with expected execution boundaries, the effectiveness of immediate decision-making can be quantified. For example, in a budget adjustment scenario, real-time feedback shows that after dynamic response solving, the average approval time decreased by 21.4%, and the frequency of abnormal jump paths decreased by 36.2%, demonstrating that dynamic solution decision-making has a significant optimization effect.

[0069] Step S6: Identify the deviation of decision execution response based on the real-time execution effect, and perform in-depth optimization of strategy iteration to build an intelligent decision-making body for internal control management.

[0070] In this embodiment, a real-time monitoring mechanism needs to be established to continuously track the execution status of each decision objective. This includes monitoring key indicators such as task completion time, resource consumption, execution order, and task success rate. Real-time monitoring continuously collects data and compares the execution results with preset objectives to identify deviations. By comparing the execution results with preset decision targets, the system can accurately determine whether deviations exist during the execution process. If the execution time of an objective exceeds a set threshold, or resource consumption exceeds expectations, the system will automatically mark these deviations. For each decision objective, the system will assess the degree of execution deviation by comparing the actual results with the expected target values. Identified deviations will be further classified and analyzed; possible sources of deviation include uneven resource allocation, incorrect task priority judgment, and unreasonable scheduling algorithms. Through in-depth analysis of the causes of deviations, the system can provide targeted solutions for subsequent optimization. Once a deviation in decision execution is identified, the system will adjust the current decision model. Adjustments may include reassessing objective priorities, reconfiguring resources, and adjusting the task execution order. The system can use a feedback control mechanism to transmit deviation information to the decision model and use this information for dynamic optimization. If a task encounters a performance bottleneck during execution, the system may adjust the task's priority or change the resource allocation strategy to ensure the overall system's efficient operation. To achieve deep optimization, the system can utilize reinforcement learning algorithms to iteratively optimize the strategy. Through a reward mechanism, the system learns continuously during execution, gradually improving its decision-making process. Whenever the system identifies a deviation, the reinforcement learning model adjusts according to the type of deviation and optimizes future decision-making strategies. In this process, the decision-making agent accumulates experience and gradually improves decision accuracy by simulating decision-making execution under different scenarios. In multi-objective decision-making, different objectives may conflict with each other, thus requiring adaptive methods to optimize the decision path for each objective. The system can dynamically adjust the weights of each objective based on factors such as importance, priority, and resource consumption, ensuring reasonable resource allocation. Through adaptive adjustment, the system can cope with changing environments and objectives, optimizing decision results. An internal control management decision-making agent framework is constructed to handle and optimize multiple decision objectives. The agent framework will include three core parts: a perception layer, a decision layer, and an execution layer. The perception layer is responsible for acquiring environmental and execution data in real time, the decision-making layer is responsible for adjusting decisions based on environmental changes, and the execution layer is responsible for actually executing decisions and providing feedback on the results. This agent possesses self-learning capabilities, extracting experience from each decision execution and gradually improving decision quality through policy iteration and reinforcement learning. Based on the real-time execution effect of each objective and feedback on deviations, the agent continuously adjusts resource allocation schemes and decision paths, gradually forming an efficient decision-making model.In multi-objective decision-making, an intelligent agent must not only solve individual objective problems but also consider the relationships between multiple objectives, balancing conflicts and resource allocation. Therefore, the agent will possess a certain degree of collaborative decision-making capability, enabling dynamic coordination among objectives to ensure optimal overall decision-making.

[0071] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include:

[0072] Collect heterogeneous raw data streams from the unit's internal control management platform; the heterogeneous raw data streams include budget management data, revenue management data, expenditure management data, procurement management data, contract management data, and materials management data;

[0073] Identify the data stream channels of the heterogeneous original data stream and perform information entropy distribution calculation to generate the information entropy density of each data stream channel;

[0074] Based on the information entropy density, spatiotemporal dimension entropy change differential analysis is performed to obtain the data entropy change evolution trajectory of each data stream channel;

[0075] Data topology analysis and modeling are performed on the heterogeneous raw data streams to construct a topology model for each type of data stream;

[0076] Based on the aforementioned topological model and the aforementioned data entropy change evolution trajectory, multi-dimensional entropy change characteristics are mined to generate a cognitive entropy change benchmark feature map.

[0077] In this embodiment, after obtaining authorization from relevant departments and users, the first step within the organization's internal control management system is to complete the unified collection of heterogeneous data sources from multiple business segments and the structured construction of heterogeneous raw data streams. The core task of this step is to break down multiple data silos, extracting data scattered across the budget management system, revenue accounting system, procurement and bidding system, contract archive system, and asset and material management platform in a streaming manner to form heterogeneous raw data streams with temporal continuity and structural consistency. These heterogeneous raw data streams constitute the underlying support for subsequent entropy change feature analysis and cognitive modeling.

[0078] First, at the data acquisition layer, a distributed acquisition tool (such as Apache NiFi, Talend, or a self-developed middleware) based on the ETL (Extract-Transform-Load) strategy is used. This is achieved through interface calls (such as RESTful APIs, JDBC connections, WebService interfaces, etc.) or direct database connections, with timed trigger rules set (e.g., acquisition every 10 minutes or when a transaction update occurs). The tool collects detailed annual and quarterly budget plans from the budget system; daily revenue records from the revenue system; each reimbursement or payment record and its approval history from the expenditure system; purchase orders, approvals, payments, and delivery records from the procurement system; contract terms, amounts, and performance progress from the contract system; and information on inbound / outbound operations, item transfers, and inventory status from the materials management system. Each type of data must include key fields such as timestamps, responsible person codes, and business process IDs for subsequent correlation modeling.

[0079] Secondly, during data standardization, it is necessary to unify the fields and map the structure of various data streams. For example, the unit of the "amount" field is unified to RMB yuan, and the format of the time field is unified to a UTC+8 timestamp (accurate to the second). The semantics of the fields are mapped through a data dictionary to avoid information fragmentation caused by different naming conventions between systems. The collected data is initially divided into structured data (such as tabular financial data), semi-structured data (such as form information extracted from scanned PDF contracts), and log data streams (such as material entry and exit logs). The structured data is stored in DataFrame format, while the unstructured data is transformed into a structured form through OCR recognition and text structure parsing.

[0080] In this phase of the experiment, a unit's integrated financial platform was selected as the pilot. The data collection period was set at 7 days, and a total of 34,587 heterogeneous raw data streams were collected, including 7,321 budget management data streams, 4,212 revenue data streams, 10,188 expenditure data streams, 9,876 procurement and contract data streams, and 2,990 material data streams. The total amount of raw data collected was approximately 1.2 GB (after compression), of which nearly 80% was structured form data, and the remainder was semi-structured text.

[0081] Ultimately, through data preprocessing and unified format conversion, a complete data foundation was laid for the next step of data flow channel identification and information entropy analysis, and multimodal raw data support was provided for the formation of a data-driven internal control model. This step solved the problems of heterogeneous data formats, scattered sources, and inconsistent time series, creating a controllable data environment for building a unified entropy change modeling system.

[0082] In this embodiment, the specific steps for mining multi-dimensional entropy change characteristics based on the topological structure model and the data entropy change evolution trajectory to generate a cognitive entropy change benchmark feature map for each data flow channel are as follows:

[0083] Based on the cognitive entropy change evolution trajectory, a cognitive response mapping is performed on the data stream topology to obtain the cognitive load lag characteristics of each channel.

[0084] The data flow complexity is calculated for the heterogeneous original data streams to obtain the complexity parameters for each type of data stream;

[0085] Based on the complexity parameters and data flow topology, instantaneous cognitive entropy mutation detection is performed to identify the cognitive state fluctuation of the data flow.

[0086] The cognitive load lag characteristics and cognitive state fluctuations are analyzed by multi-dimensional entropy change characteristics of the data stream, and a cognitive entropy change benchmark feature map of each data stream channel is generated.

[0087] In this embodiment, after modeling the information entropy evolution trajectory, the temporal variation pattern of the data stream has been quantified into a continuous entropy change curve. The core of this stage is to analyze the response order and reaction delay between different channels to changes in system state, thereby identifying the "cognitive response capability" of each channel. The strength of this capability can be quantified by the "cognitive load lag characteristic." Specifically, this means determining the time lag in which a channel's information entropy changes significantly when facing system mutations or interference. In the implementation process, it is necessary to compare the entropy change curves of different data channels and use cross-correlation analysis to identify the time delay relationship between each channel. The information entropy sequence of each channel is divided into equal-width time windows, and the maximum correlation delay point of the entropy value sequence between each pair of channels is calculated. If a channel has a continuous time delay response compared to its upstream node, that is, its entropy value mutation occurs after the downstream node, then the channel is considered to have a high cognitive lag. In the entire network data topology, the lag relationship of all channels can form a lag mapping matrix. By statistically aggregating the matrix, the average lag time, lag variance, minimum and maximum response delay, and other indicators of each channel can be calculated. These metrics collectively define the cognitive load response curve of a channel. If the response lag value of a channel is significantly higher than that of its topological neighbors, it indicates that it is experiencing perceptual lag in the event chain, facing high processing pressure and exhibiting weak system resilience. In dynamic monitoring, identifying high-lag channels helps to discover potential information bottlenecks and propagation delay risks in advance, and optimize flow investigation strategies and model scheduling. The data structure, change patterns, and predictability of each data channel are quantified, defining its complexity parameters. Complexity is an important indicator for measuring the structure and uncertainty of a data sequence, reflecting the processing difficulty and resource consumption that the data brings to the cognitive model. Generally, higher complexity indicates that the data channel is more irregular, changes more frequently, and the model finds it more difficult to accurately model its state. Time-series features of the data stream are extracted, and basic statistics are calculated, such as volatility (variance), rate of change (mean of first-order differences), and periodicity (peak value of the autocorrelation function). These features constitute the basic indicators of time complexity. Data structure complexity is evaluated, including factors such as the number of fields, nesting levels, and data format consistency. Structured data typically has lower complexity compared to semi-structured data, but the real-time processing requirements of high-frequency data can lead to increased cognitive complexity. Furthermore, methods based on approximate entropy or sample entropy can be introduced to model the predictability of data streams. These methods assess the compressibility and pattern repeatability of sequences by comparing the similarity between different subsequences. The more incompressible and irregular the sequence, the higher its approximate entropy, representing greater complexity. Normalizing these various complexity metrics forms a unified complexity parameter vector, representing the overall complexity of each channel. This parameter is used not only for model input features but also for subsequent entropy mutation detection to determine whether the change is "structural" or "abnormal."The goal is to determine whether the data stream experienced a "sudden" state change at a specific point in time or within a time window, i.e., to identify so-called "instantaneous cognitive entropy mutations." Unlike the evolutionary analysis in the previous stage, this stage focuses on the phenomenon of drastic fluctuations in entropy values ​​over a short period of time, especially those changes that cannot be explained by normal complexity.

[0088] The implementation process consists of three core steps. First, monitor the change in information entropy of each data channel over time, calculating the increase or decrease in entropy using a sliding time window method. If the entropy difference between two consecutive windows exceeds a set threshold and shows a continuous upward (or downward) trend, it is initially judged as a mutation signal. Second, compare the detected mutation points with the complexity parameters of the channel. If the data itself has high complexity (such as high-frequency noise or strong periodicity), the mutation may be "structural noise" and should not be considered an anomaly; however, when the mutation occurs in a low-complexity channel, it is more likely to be a real abnormal behavior. This joint analysis helps filter out false anomalies and improves detection accuracy. Third, combine the topology of the data flow to analyze whether the mutation occurs simultaneously or successively in multiple related channels, thereby determining its propagation characteristics. If the mutation affects multiple channels simultaneously or propagates step by step along a certain path, it indicates that the mutation has systemic influence and is an important input signal for the cognitive model. By using multiple indicators such as entropy mutation frequency, intensity, and propagation path length, a "cognitive state fluctuation index" can be constructed to measure the stability and disturbance intensity of the current system, providing a decision-making basis for subsequent anomaly response strategies. The extracted features are integrated and modeled to form a complete entropy change feature profile, constructing a "cognitive entropy change baseline feature map" for each data channel. This map serves as a structured input for anomaly detection models, scheduling systems, and optimization algorithms, providing a complete and traceable model of data cognitive behavior. The map generation involves four stages: feature aggregation, normalization, structural mapping, and visual representation. The entropy statistics (such as average entropy, variance, kurtosis), entropy mutation indices (frequency, intensity), complexity parameters (structure, compressibility), topological features (in-degree, out-degree, betweenness), and cognitive lag indices (average response delay, degree of variation) of each channel are all aggregated into a unified vector space, forming a multi-dimensional feature vector. Standardization methods are used to scale all features, ensuring comparability across different dimensions. Then, dimensionality reduction methods (such as principal component analysis or multi-scale embedding) are used to map high-dimensional features to 2D or 3D space for cluster analysis and map visualization. This process categorizes different types of data streams into several classes, such as high-entropy-high-complexity, low-entropy-high-response, and mutation-diffusion types. Each channel corresponds to a complete graph node, whose attributes describe its behavioral characteristics, mutation risk level, propagation role in the network, and response characteristics. This graph can be used not only for offline model analysis but also for quickly screening key monitoring channels and identifying potentially high-risk nodes in real-time systems, serving as an important reference for AI system optimization scheduling and response prioritization.

[0089] In this embodiment, see Figure 3 The diagram below illustrates the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include:

[0090] Multi-level semantic parsing is performed on the heterogeneous raw data stream to obtain semantic anomaly features, temporal anomaly fluctuations, and data correlation anomaly features.

[0091] Adaptive semantic filtering enhancement is applied to semantic anomaly features to obtain filtered and optimized anomaly semantic feature data.

[0092] Multi-dimensional semantic space projection is performed on the filtered and optimized abnormal semantic feature data to obtain the multi-level semantic space of each data stream;

[0093] Based on the abnormal fluctuations in time series and the abnormal correlations between data, the multi-level semantic space is subjected to deep semantic analysis to extract the deep semantic kernel and surface semantic representation of the abnormal patterns.

[0094] The deep semantic kernel and surface semantic representation are subjected to abnormal behavior semantic mining, and the abnormal pattern semantics are deconstructed layer by layer to construct a precise semantic feature profile for each abnormal pattern.

[0095] In this embodiment, within the unit's multimodal data-driven internal control decision analysis system, the heterogeneous data streams generated by different business systems (such as budgets, revenues, contracts, procurement, and materials) possess highly diverse structural and semantic characteristics. Therefore, before performing decision modeling, it is essential to systematically analyze the semantics and model the abnormal features of these data. The goal of this step is to extract abnormal patterns from the complex raw data through multi-level semantic analysis and adaptive processing, and to construct a "precise semantic feature profile" with deep logical interpretability, providing an interpretable basis for subsequent anomaly identification and risk response.

[0096] The first stage is multi-level semantic parsing. Based on Natural Language Processing (NLP) technology, Graph Neural Networks (GNNs), and structured data parsing algorithms, the system hierarchically decomposes text fields (such as contract terms and procurement instructions), structured tabular fields (such as budget entries and expenditure details), and time-series data (such as procurement frequency and budget inflow time series) from heterogeneous data. By constructing a multi-source semantic vector embedding model, structured and unstructured information are uniformly mapped into the semantic space. During the parsing process, three types of key anomalous features are extracted: first, semantic anomalous features (such as inconsistencies between budget usage and contract terms); second, anomalous fluctuations in time series (such as expenditures occurring earlier or later than the budget approval cycle); and third, anomalous features in data correlation (such as a decrease in the logical alignment between budget items and procurement plans). In experiments, the embedding model based on the Transformer architecture successfully identified over 92% of semantic deviation samples in the analysis of a unit's budget-procurement data flow in 2023.

[0097] The second stage is adaptive semantic filtering enhancement. Due to the large amount of noise and ambiguous semantic expressions in the original data, the system needs to utilize a self-attention-based filtering enhancement model to screen and enhance the initially extracted semantic anomaly features. In practical applications, a semantic sensitivity factor (SSF) is introduced to measure the degree of influence of semantic variations on internal control rules. The model reconstructs the data by weighting the impact of each type of anomalous semantic on the control objective, retaining high-weight anomalies and weakening low-relevance semantics, forming a set of "filtered and optimized anomalous semantic feature data".

[0098] The third stage is multi-dimensional semantic space projection. The system uses t-SNE dimensionality reduction and Principal Component Analysis (PCA) to map the filtered semantic features to different semantic space dimensions. Each dimension represents the semantic mapping result of a specific business entity (such as budget items, contract types, procurement processes, etc.). Taking the "budget procurement semantic space" as an example, dimension one represents "budget intent," dimension two represents "procurement path legitimacy," and dimension three represents "approval chain dependency." These multi-dimensional projections form a visualized semantic network, providing a logical structural foundation for further deconstruction of abnormal patterns.

[0099] The fourth stage is deep semantic analysis. Building upon the existing multidimensional semantic space, and combining the previously identified "abnormal fluctuations in time series" and "abnormal features of data correlation," a multimodal fusion model (such as Cross-Modal BERT) is used to extract the deep semantic kernel (i.e., the logical factors that truly trigger abnormal behavior) and the surface semantic representation (i.e., the semantic phenomena that are perceptible to the outside world) of each abnormal pattern. For example, in contract performance data, "the use of funds deviates from the contract description" is identified. Its deep semantic kernel is "deviation in the implementation of project expenditure classification standards," and its surface semantic is "inconsistency between the expenditure item name and the contract."

[0100] Finally, the system enters the stage of abnormal behavior semantic mining and layer-by-layer semantic deconstruction. Based on graph-semantic mining and a knowledge rule engine, the system performs semantic path backtracking, rule conflict detection, and behavior pattern clustering on the combination relationship between deep and surface semantics. By constructing an "abnormal semantic decision tree," each type of abnormal pattern is deconstructed layer by layer, ultimately generating a "precise semantic feature profile" that includes semantic triggering conditions, data feature range, abnormal behavior path, and possible business consequences.

[0101] In this embodiment, see Figure 4 The diagram below illustrates the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include:

[0102] Obtain a preset decision target instruction; identify multiple decision targets based on the decision target instruction; calculate the real-time weight changes of the decision targets;

[0103] Based on the multi-decision objective, the real-time target conflict intensity is calculated by filtering and optimizing the abnormal semantic feature data.

[0104] A deep analysis of the conflict situation correlation is performed on the real-time weight changes and real-time target conflict intensity to obtain the conflict situation correlation curve;

[0105] Based on the cognitive entropy change baseline feature map and the conflict situation correlation curve, multi-objective coordination request analysis is performed to obtain the coordination request signal for each decision objective.

[0106] The coordination request signal is dynamically evaluated to obtain the coordination request priority of each decision objective.

[0107] In this embodiment, within the organization's internal control system, different decision-making objectives often overlap and conflict in terms of resources, time, etc. For example, between objectives such as budget balancing, compliance assurance, efficiency improvement, and procurement execution rate, the decision-making side needs to dynamically adjust its trade-off strategies in a time-varying environment. Therefore, this step aims to achieve real-time analysis and coordination optimization of multi-objective conflict situations through a systematic conflict modeling and priority assessment method, ensuring that internal control behavior has dynamic adaptability and precise execution capabilities. The system obtains preset decision-making objective instructions from the internal control platform. These instructions are usually predefined by the upper management and include specific objectives (such as "the quarterly budget execution rate shall not be lower than 85%" and "contract performance risk shall be controlled within 3%) and their initial priorities. The system constructs a target mapping matrix based on rule labels, indicator fields, and strategy vectors, and uniformly abstracts and models each decision-making objective in a parameterized form. The system parses the instructions and identifies multiple sub-decision-making objectives with mutually exclusive, competitive, or dependent relationships. This process, based on an ontology recognition model and a historical behavioral corpus training set, infers a structural relationship graph between objectives through semantic relationships. Taking annual budget and procurement data as an example, structural modeling identifies 12 types of target combinations with typical conflict attributes, such as the clear inverse coupling relationship between "timely supply rate of materials" and "procurement cost saving rate". The system introduces a "real-time weight change calculation mechanism" to adjust the weights of each target based on the dynamic changes of real-time data in the business environment (such as material shortages, contract delays, approval backlogs, etc.). Here, a dynamic evaluation model based on Temporal Causal Bayesian Network (TCBN) is used to evaluate the relative strength of the target's impact on organizational operations in real time. For example, during emergencies, the weight of "emergency procurement completion rate" increases significantly due to its criticality to ensuring organizational operation. For the state of multiple targets existing in parallel, the system matches filtered and optimized abnormal semantic feature data with a multi-target behavioral demand model to calculate the real-time conflict intensity between each target. By constructing a conflict mapping tensor, the system measures the degree of conflict overlap in the intersection area of ​​different target resources in the current data scenario and visualizes the conflict intensity value in the form of a heatmap.

[0108] Building upon this foundation, the system conducts in-depth analysis of the conflict situation correlation between "real-time weight changes" and "target conflict intensity," constructing a Conflict Situation Association Curve. This curve, through time-series modeling and multi-target evolutionary graph analysis, depicts the dynamic evolution trend of conflict relationships between targets over time, providing a basis for scheduling decisions. For example, it reveals periodic fluctuations in some target conflicts, corresponding to the end-of-quarter budget sprint period or the concentrated contract signing period at the beginning of the year. At the cognitive level, the system jointly models the cognitive entropy change baseline feature graph with the aforementioned conflict situation curve to conduct multi-target coordination request analysis. This step utilizes a self-attention mechanism to jointly model the semantic shift trend, conflict source dimensions, and risk spillover paths of each target, generating corresponding coordination request signals (such as increasing target flexibility, allowing for indicator variation ranges, and calling external resources for compensation).

[0109] The system comprehensively evaluates the coordination request priority of each objective based on objective weight, conflict intensity, coordination cost, and strategy impact. The priority evaluation model, based on fuzzy logic and multilayer perceptron (MLP), dynamically outputs objective coordination priority scores and performs ranking and scheduling at the control center.

[0110] In this embodiment, step S4 includes the following steps:

[0111] The maximum decision capacity of the cognitive entropy change baseline feature map is calculated based on the multiple decision objectives to obtain the maximum carrying capacity of each decision objective.

[0112] Based on the accurate semantic feature profile, a multi-time point decision load assessment is performed to obtain a multi-time point decision load assessment curve.

[0113] Multi-scenario decision-making simulations were performed on the maximum carrying capacity and multi-time point decision load evaluation curves to obtain multiple scenario decision-making simulation data.

[0114] Based on the coordination request priority, dynamic optimal decision boundaries are calculated for multiple scenario decision simulation data, thereby generating a dynamic trade-off constraint range for each decision objective.

[0115] In this embodiment, in multi-objective optimization decision-making, each decision objective has different requirements for system resources and varying impacts on the system. Therefore, evaluating the "maximum carrying capacity" of each decision objective is crucial. This carrying capacity reflects the maximum extent to which a certain objective can be effectively executed under the current system state. Specifically, the system analyzes the entropy change characteristics of each data channel in the "cognitive entropy change baseline feature map" to evaluate the resource consumption and impact range of each objective under specific data characteristics. A certain objective may exhibit higher sensitivity in high entropy change regions, while potentially leading to resource waste in low entropy change regions. Therefore, by analyzing these characteristics, the system can calculate the maximum carrying capacity of each decision objective. In the specific calculation:

[0116] Input parameters:

[0117] Each decision objective (such as anomaly classification, resource scheduling, and fault prediction) has a set of corresponding entropy change feature sensitive dimensions;

[0118] The entropy structure of different channels in the spectrum characterizes their sensing load capacity and signal transparency;

[0119] Each channel has known metrics such as volatility, lag, and maximum entropy.

[0120] Based on the information theory model and the system entropy budget model, the maximum decision capacity is defined as:

[0121] ;

[0122] Where, α_j^((i)): the weight of the i-th target on the j-th channel;

[0123] V_j: Volatility of channel j;

[0124] L_j: The response lag of channel j;

[0125] E_j^max: Maximum information entropy that the channel can accommodate;

[0126] The model is solved using linear programming constraints; a maximum acceptable load is set for each target, and the feature map value of each channel in the current state is normalized. The final output is a maximum carrying capacity C_max^((i)) for each target.

[0127] Over time, the system's operating state and external environment may change, affecting the execution effectiveness of various decision objectives. Therefore, evaluating the changes in decision load at multiple time points is crucial for dynamic decision optimization. The system analyzes the performance of each objective at different time points in the "precise semantic feature profile" and combines this with the characteristics of real-time data streams to calculate the load of each objective at each time point. These evaluation results are plotted as "multi-time point decision load evaluation curves" to reflect the load change trends of each objective over different time periods. In experiments, the system typically sets multiple time windows (e.g., 10 seconds, 30 seconds, 60 seconds, etc.) and calculates the load of each objective within each window. By comparing the evaluation results of different time windows, the system can identify patterns in load changes, providing a basis for subsequent decision optimization. In practical applications, the system may face various operating scenarios, such as high load, low latency, and large data volumes. To cope with these variable scenarios, the system needs to perform multi-scenario decision simulations. By using the "maximum carrying capacity" and the "multi-time point decision load evaluation curves" as inputs, the system can construct multiple scenario models to simulate the performance of each decision objective under different scenarios. These simulation results, termed "scenario-based decision simulation data," are used to evaluate the system's decision-making effectiveness and resource utilization under various conditions. In experiments, the system typically sets up multiple simulation scenarios (such as high-load scenarios, low-latency scenarios, and large-data-volume scenarios) and performs decision simulations in each scenario. By comparing the simulation results across different scenarios, the system can identify the optimal decision-making strategy, providing guidance for practical applications. In multi-objective optimization decision-making, conflicts or competition may exist between objectives; effectively balancing these objectives is a key issue. The system analyzes "coordination request priorities" and combines this with "scenario-based decision simulation data" to calculate the optimal decision boundaries for each objective under different scenarios. These boundaries reflect the best state achievable by each objective under specific conditions. Through the analysis of these boundaries, the system can determine the "dynamic trade-off constraint range" for each decision objective, i.e., the optimal range acceptable to each objective in the current environment. The system typically uses multi-objective optimization algorithms, such as genetic algorithms and particle swarm optimization, combined with real-world data for calculation. By comparing the results of different algorithms, the system can select the decision-making strategy most suitable for the current application scenario.

[0128] In this embodiment, step S5 includes the following steps:

[0129] Multi-objective dynamic constraint response is solved based on the dynamic trade-off constraint range to generate multiple objective decision paths;

[0130] Risk path extrapolation analysis is performed on multiple target decision paths to obtain the risk extrapolation characteristics of each path;

[0131] Based on the risk simulation characteristics, adaptive decision-making path planning is performed to construct a multi-objective collaborative decision-making execution strategy;

[0132] Define real-time control instructions based on multi-objective collaborative decision-making and execution strategies, and generate multiple decision-objective execution control instructions;

[0133] Based on the decision objectives, control commands are executed to perform real-time decision processing, and the real-time execution effect of each decision objective is collected.

[0134] In this embodiment, the dynamic trade-off constraint range is transformed into a set of constraint boundary parameters, corresponding to the carrying capacity limit, preference direction, critical threshold, etc. of each objective. Based on these constraints, the system uses combinatorial optimization strategies (such as heuristic search, rule-matching path compression, weight adjustment sliding window, etc.) to search for possible optimal or suboptimal paths in the policy space. In experiments, the search depth for path generation is typically set to 5-7 layers, allowing dynamic jumps in objective priority between nodes. Each path corresponds to a set of policy execution order, impact range, and expected benefits. Each generated path logically satisfies the current system resource constraints and the controllability of objective conflicts, while possessing execution stability and predictability. The output path set includes not only the execution order of each objective but also dynamic constraint offsets, weight adjustment coefficients, and estimated execution resource proportions, facilitating subsequent deduction and scheduling. After path generation, the system needs to evaluate the potential uncertainties, failure probability, and anomaly probability of these paths during execution. This step uses a "risk path deduction" mechanism to simulate and conduct emergency assessments of the execution process of each candidate path, thereby identifying its risk characteristics. The simulation analysis relies on historical anomaly databases, entropy change pattern evolution trends, and system sensitivity indicators for each node in the path. In a simulation environment, the system performs multiple rounds of "time-segment" simulations for each path, breaking down the path execution into multiple time periods and evaluating the system response value (such as latency, abrupt changes, and conflict levels) for each segment. In experimental deployment, each path is simulated for 10-20 rounds, generating a set of risk characteristic data for each round, including but not limited to: Risk Exposure Point (RBP), Fluctuation Response Intensity (FRI), Execution Time Offset (DTA), and Resource Overrun Frequency (ORF). This data is aggregated and modeled to form a "risk simulation feature set" for the path. This feature set is used for subsequent path selection and strategy planning and can be compared with existing system robustness assessment models to identify high-risk and robust paths. Paths with lower risk are more likely to be used for high-priority objectives.

[0135] After understanding the risk characteristics of each path, the system enters the "path selection and combination planning" stage. The goal of this step is to select the optimal or most adaptive path combination based on the simulation results, thereby constructing a decision-making strategy that can be collaboratively executed across multiple objectives. The system first matches the risk characteristics of each path with the objective requirements, prioritizing paths with low risk exposure, low redundancy, and balanced resource utilization. Subsequently, based on the synergy between objectives (such as non-conflict or gain effects), it attempts to combine several paths into a "collaborative execution group." Collaborative path groups can simultaneously satisfy multiple objectives while sharing some computing resources, improving decision-making efficiency. In experiments, the path planning module typically employs adaptive strategy combination algorithms (such as Pareto boundary heuristic fusion, dynamic population evolution strategies, etc.) to evaluate the value of path combinations. Evaluation criteria include, but are not limited to, average response time, resource utilization improvement, and redundancy / conflict reduction. The output is a set of "multi-objective collaborative decision-making execution strategies," each strategy defining the set of objectives, priority ranking, expected duration, desired output, and fallback strategy. This strategy set is directly input into the next stage control instruction module. Abstract execution strategies are transformed into control instructions that can be directly invoked within the system, thereby driving each processing module to make real-time adjustments to data flow, model resources, and scheduling behavior. Each set of collaborative decision-making strategies is translated into a set of structured control instructions, including but not limited to: target identifiers, resource quotas, execution thresholds, feedback mechanisms, fault tolerance time windows, and rollback path instructions. The system automatically arranges the scheduling order and activation timing of control signals based on the priority allocation and load prediction in the strategies of each target. In actual deployment, each control instruction is encoded into a standard operation template (such as a JSON structured instruction package) and distributed to each computing node, analysis model, response module, or alarm system via the scheduling bus. Upon receiving the instruction, each node immediately adjusts its operating parameters, such as dynamically switching data filtering accuracy, enhancing the model anomaly capture rate, and shortening alarm latency. This process needs to consider instruction redundancy control to ensure that system load does not surge due to policy overlap. Experiments recommend that each batch of instructions not exceed 20, each instruction have a maximum execution cycle of 30 seconds, and interrupt and re-call interfaces are provided to ensure controllability and rollback. After the control instructions are issued, the system enters the "real-time decision processing" stage. During this phase, each module of the system executes tasks such as data processing, model response, and strategy adjustment according to instructions, while simultaneously collecting the execution status and performance metrics of each objective in real time. The system is equipped with a data monitoring and feedback mechanism responsible for tracking the entire execution process. Key parameters collected include: response time, execution success rate, data throughput changes, anomaly detection accuracy, alarm timeliness, and resource utilization changes. The execution performance of each objective is continuously recorded in the "Execution Performance Record Library" and archived hierarchically by timestamp for subsequent optimization strategy adjustments and reinforcement learning feedback.In the experiment, it is recommended to sample once per second, with no fewer than 10 data fields recorded for each target, ensuring that the sampling dimensions cover the four major categories of indicators: model, data, system, and resources. Furthermore, the system has an anomaly response trigger mechanism. Once the execution of a target deviates from the expected strategy by more than a set threshold (e.g., response time deviation greater than 20%), a rollback request or strategy reassessment signal will be immediately generated, triggering a subsequent adaptive adjustment mechanism to maintain the system's stability and agility.

[0136] In this embodiment, step S6 includes the following steps:

[0137] Identify the benchmark execution effect of decision-making objectives based on preset decision-making objective instructions;

[0138] Based on the benchmark execution effect of the decision target, identify the decision execution response deviation in the real-time execution effect of each decision target, and mark the decision target with execution deviation;

[0139] Conduct in-depth analysis of the causes of deviations in the execution of decision-making objectives to identify the root causes of decision failures;

[0140] Based on the root causes of decision failure, the multi-objective collaborative decision-making execution strategy is dynamically reconstructed by weighting the multi-objectives and reconstructing the decision-making strategy to obtain globally optimized decision parameters.

[0141] The global optimization decision parameters are deeply optimized through strategy iteration to construct an intelligent agent for internal control management decision-making.

[0142] In this embodiment, during the multi-objective optimization decision-making process, it is first necessary to define the baseline execution effect for each decision objective to facilitate subsequent deviation identification and optimization. The core task of this process is to quantitatively evaluate the initial execution effect of each decision objective through preset decision objective instructions. By setting standardized target performance indicators, such as response time, resource consumption, and execution success rate, the system can accurately identify the expected execution effect of each objective. Specifically, the system extracts the execution standard for each objective from the decision objective instructions. These standards are typically set based on historical data, business requirements, or expert advice. If the execution standard for a certain objective is "response time completed within 30 seconds," the system will monitor and record the actual response time for each execution. By comparing this with the preset standard value, the system generates the baseline execution effect for each objective, which is then used for subsequent execution deviation identification. In experiments, these baseline execution effects are collected and verified in multiple scenarios to ensure their adaptability and accuracy. Each objective is executed in a standard scenario, and its performance is compared to obtain the average value as a baseline. The standard value typically stabilizes within 20 to 50 experimental cycles to ensure representativeness. Once the baseline performance is determined, the system monitors the execution of the decision-making objectives in real time, identifying deviations between the actual execution and the baseline performance. This process primarily relies on real-time data monitoring and comparative analysis, using these deviations to mark decision-making objectives whose performance does not meet expectations.

[0143] The system employs a real-time execution monitoring mechanism to dynamically track and collect real-time data for each decision objective during execution. For a real-time response objective, the system monitors key indicators such as response time, execution completion rate, and resource consumption. The real-time execution performance is compared with a preset benchmark performance; any deviation from a certain threshold is marked as an "execution deviation." In experiments, a threshold range (e.g., ±10% deviation) is typically set. Once the target execution deviation exceeds this range, the system immediately records and marks the relevant decision objective as a "deviation objective." This helps to promptly identify problems and initiate further analysis. Through continuous tracking of these deviation objectives, the system can accumulate a large amount of execution deviation data, providing data support for subsequent in-depth analysis of the causes of deviations. In-depth analysis of the causes of deviations is a crucial step in this method, aiming to identify the root causes of execution deviations for targeted optimization. The causes of deviations may originate from multiple levels, such as data anomalies, insufficient resource allocation, and inaccurate algorithm models. The system uses multi-dimensional analysis techniques to conduct in-depth analysis of deviation objectives. The system traces the target's execution path to analyze whether changes in the external environment (sudden fluctuations in data flow or drastic changes in system resources) have affected the target's execution. Secondly, the system combines historical execution records, log data, and model outputs to identify potential defects in the algorithm or model during execution. The system also compares the target's execution pattern with the execution patterns of other normal targets to identify potential differences and conduct in-depth analysis.

[0144] In experiments, the system often uses causal inference models or feature importance analysis methods from machine learning to help identify the root causes of deviations. Through feature selection and dimensionality reduction techniques, the system can quickly identify key factors related to decision failure, such as the impact of changes in certain external parameters on target execution or data input quality issues. After identifying the root causes of decision failure, the system enters the "decision strategy reconstruction" stage. The core task of this stage is to adjust the existing multi-objective collaborative decision-making strategy based on the identified failure factors, optimizing its target weights and decision execution process. The system first quantifies and evaluates the key factors affecting decision failure. If abnormal fluctuations in the data flow are the main factor causing execution deviations, the system will increase the influence of this factor on the decision by adjusting the weight values ​​of the corresponding targets. Furthermore, the system can re-plan the decision strategy according to the actual situation, achieving global optimization by adjusting the order of target execution and the priority of resource allocation. In experiments, methods such as genetic algorithms and particle swarm optimization are typically used to dynamically reconstruct the decision strategy, simulating the execution effects of different strategies to select the optimal strategy combination. After each strategy optimization, the system verifies through backtesting and forward testing whether the new weight settings and execution strategies effectively reduce execution bias and improve the execution effect of the objectives. The reconstructed decision strategy will include the adjusted weights of each objective, a new resource scheduling plan, and execution priorities, ensuring that the entire system achieves more efficient collaboration and optimization among multiple objectives. After obtaining the global optimization decision parameters, the system enters the final strategy iteration optimization stage. The purpose of this stage is to further optimize the performance of the decision agent through continuous strategy iteration, enabling it to execute multi-objective optimization decisions more efficiently in future real-time data streams and complex scenarios. The strategy iteration process typically includes two main steps: Based on the current optimization decision parameters, the system conducts multiple rounds of simulation and testing to evaluate the execution effect of the strategy. By collecting various performance indicators during the execution process, the system further adjusts the strategy. Secondly, the system employs intelligent methods such as reinforcement learning, enabling the decision agent to learn autonomously and gradually optimize its decision-making process. This feedback-based iterative optimization ensures that the system can continuously improve and adapt to new situations when facing complex and dynamically changing environments. In experiments, a certain number of iterations (e.g., 10, 20, etc.) is typically set, and the robustness and adaptability of the strategy are tested through a dynamically changing data environment. Feedback information after each iteration is used to adjust the agent's decision-making behavior, further improving its performance in unknown environments. After multiple policy optimizations, the system forms a highly intelligent, adaptively adjusting internal control management decision-making agent. This agent can continuously play a role in complex data flow anomaly perception and optimization decision-making tasks.

[0145] In this embodiment, a unit internal control management decision-making system based on multimodal data analysis is provided for executing the unit internal control management decision-making method based on multimodal data analysis as described above, including:

[0146] The data stream entropy change analysis module is used to collect heterogeneous raw data streams from the unit's internal control management platform; perform spatiotemporal entropy change differential analysis and multi-dimensional entropy change characteristic mining to generate a cognitive entropy change benchmark feature map;

[0147] The semantic parsing module is used to perform multi-level semantic parsing on the heterogeneous raw data stream, and to perform layer-by-layer abnormal pattern semantic deconstruction to construct a precise semantic feature profile for each abnormal pattern.

[0148] The multi-objective coordination module is used to obtain preset decision-making objective instructions, perform multi-objective coordination request analysis and dynamic objective priority evaluation on the cognitive entropy change benchmark feature map, and obtain the coordination request priority.

[0149] The decision boundary module is used to calculate the maximum decision capacity of the cognitive entropy change benchmark feature map, and to calculate the dynamic optimal decision boundary based on the coordination request priority and accurate semantic feature profile, thereby generating a dynamic trade-off constraint range.

[0150] The constraint solving module is used to solve multi-objective dynamic constraint responses based on the dynamic trade-off constraint range, and to perform real-time decision processing to obtain real-time execution results.

[0151] The internal control management decision module is used to identify deviations in decision execution response based on real-time execution results, and to perform in-depth optimization of strategy iteration to build an intelligent internal control management decision-making body.

[0152] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0153] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A unit internal control management decision-making method based on multimodal data analysis, characterized in that, Includes the following steps: Step S1: Collect heterogeneous raw data streams from the unit's internal control management platform; perform spatiotemporal entropy change differential analysis and multi-dimensional entropy change characteristic mining to generate a cognitive entropy change benchmark feature map; Step S2: Perform multi-level semantic parsing on the heterogeneous raw data stream, and perform layer-by-layer abnormal pattern semantic deconstruction to construct a precise semantic feature profile for each abnormal pattern; Step S3: Obtain the preset decision target instruction, perform multi-target coordination request analysis and dynamic target priority evaluation on the cognitive entropy change benchmark feature map, and obtain the coordination request priority; Step S4: Calculate the maximum decision capacity of the cognitive entropy change baseline feature map, and calculate the dynamic optimal decision boundary based on the coordination request priority and accurate semantic feature profile, thereby generating a dynamic trade-off constraint range. Step S5: Solve the multi-objective dynamic constraint response based on the dynamic trade-off constraint range, and perform real-time decision processing to obtain the real-time execution effect; Step S6: Identify the deviation of decision execution response based on the real-time execution effect, perform in-depth optimization of strategy iteration, and build an intelligent decision-making body for internal control management; The specific steps of step S1 are as follows: Collect heterogeneous raw data streams from the unit's internal control management platform; the heterogeneous raw data streams include budget management data, revenue management data, expenditure management data, procurement management data, contract management data, and materials management data; Identify the data stream channels of the heterogeneous original data stream and perform information entropy distribution calculation to generate the information entropy density of each data stream channel; Based on the information entropy density, spatiotemporal dimension entropy change differential analysis is performed to obtain the data entropy change evolution trajectory of each data stream channel; Data topology analysis and modeling are performed on the heterogeneous raw data streams to construct a topology model for each type of data stream; Based on the aforementioned topological structure model and the aforementioned data entropy change evolution trajectory, multi-dimensional entropy change characteristics are mined to generate a cognitive entropy change benchmark feature map. The specific steps for mining multi-dimensional entropy change characteristics based on the topological structure model and the data entropy change evolution trajectory to generate a cognitive entropy change benchmark feature map are as follows: Based on the cognitive entropy change evolution trajectory, a cognitive response mapping is performed on the data stream topology to obtain the cognitive load lag characteristics of each channel. The data flow complexity is calculated for the heterogeneous original data streams to obtain the complexity parameters for each type of data stream; Based on the complexity parameters and data flow topology, instantaneous cognitive entropy mutation detection is performed to identify the cognitive state fluctuation of the data flow. The cognitive load lag characteristics and cognitive state fluctuations are analyzed using multi-dimensional entropy change characteristics of the data stream to generate a cognitive entropy change baseline feature map.

2. The unit internal control management decision-making method based on multimodal data analysis according to claim 1, characterized in that, The specific steps of step S2 are as follows: Multi-level semantic parsing is performed on the heterogeneous raw data stream to obtain semantic anomaly features, temporal anomaly fluctuations, and data correlation anomaly features. Adaptive semantic filtering enhancement is applied to semantic anomaly features to obtain filtered and optimized anomaly semantic feature data. Multi-dimensional semantic space projection is performed on the filtered and optimized abnormal semantic feature data to obtain the multi-level semantic space of each data stream; Based on the abnormal fluctuations in time series and the abnormal correlations between data, the multi-level semantic space is subjected to deep semantic analysis to extract the deep semantic kernel and surface semantic representation of the abnormal patterns. The deep semantic kernel and surface semantic representation are subjected to abnormal behavior semantic mining, and the abnormal pattern semantics are deconstructed layer by layer to construct a precise semantic feature profile for each abnormal pattern.

3. The unit internal control management decision-making method based on multimodal data analysis according to claim 1, characterized in that, Step S3 is as follows: Obtain a preset decision target instruction; identify multiple decision targets based on the decision target instruction; calculate the real-time weight changes of the decision targets; Based on the aforementioned multi-decision objectives, the real-time target conflict intensity is calculated by filtering and optimizing the abnormal semantic feature data to obtain the real-time target conflict intensity. A deep analysis of the conflict situation correlation is performed on the real-time weight changes and real-time target conflict intensity to obtain the conflict situation correlation curve; Based on the cognitive entropy change baseline feature map and conflict situation correlation curve, multi-objective coordination request analysis is performed to obtain the coordination request signal for each decision objective. The coordination request signal is dynamically evaluated to obtain the coordination request priority of each decision objective.

4. The unit internal control management decision-making method based on multimodal data analysis according to claim 3, characterized in that, The specific steps of step S4 are as follows: The maximum decision capacity of the cognitive entropy change baseline feature map is calculated based on the multiple decision objectives to obtain the maximum carrying capacity of each decision objective. Based on the accurate semantic feature profile, a multi-time point decision load assessment is performed to obtain a multi-time point decision load assessment curve. Multi-scenario decision-making simulations were performed on the maximum carrying capacity and multi-time point decision load evaluation curves to obtain multiple scenario decision-making simulation data. Based on the coordination request priority, dynamic optimal decision boundaries are calculated for multiple scenario decision simulation data, thereby generating a dynamic trade-off constraint range for each decision objective.

5. The unit internal control management decision-making method based on multimodal data analysis according to claim 1, characterized in that, The specific steps of step S5 are as follows: Multi-objective dynamic constraint response is solved based on the dynamic trade-off constraint range to generate multiple objective decision paths; Risk path extrapolation analysis is performed on multiple target decision paths to obtain the risk extrapolation characteristics of each path; Based on the risk simulation characteristics, adaptive decision-making path planning is performed to construct a multi-objective collaborative decision-making execution strategy; Define real-time control instructions based on multi-objective collaborative decision-making and execution strategies, and generate multiple decision-objective execution control instructions. Based on the decision objectives, control commands are executed to perform real-time decision processing, and the real-time execution effect of each decision objective is collected.

6. The unit internal control management decision-making method based on multimodal data analysis according to claim 1, characterized in that, The specific steps of step S6 are as follows: Identify the benchmark execution effect of decision-making objectives based on preset decision-making objective instructions; Based on the benchmark execution effect of the decision target, identify the decision execution response deviation of each decision target in real time, and mark the decision target with execution deviation; Conduct in-depth analysis of the causes of deviations in the decision-making objectives to identify the root causes of decision failures; Based on the root causes of decision failure, the multi-objective collaborative decision-making execution strategy is dynamically reconstructed by weighting the multi-objectives and reconstructing the decision-making strategy to obtain globally optimized decision parameters. The global optimization decision parameters are deeply optimized through strategy iteration to construct an intelligent agent for internal control management decision-making.

7. A unit internal control management decision-making system based on multimodal data analysis, characterized in that, The method for implementing the unit internal control management decision-making method based on multimodal data analysis as described in claim 1 includes: The data stream entropy change analysis module is used to collect heterogeneous raw data streams from the unit's internal control management platform; perform spatiotemporal entropy change differential analysis and multi-dimensional entropy change characteristic mining to generate a cognitive entropy change benchmark feature map; The semantic parsing module is used to perform multi-level semantic parsing on the heterogeneous raw data stream, and to perform layer-by-layer abnormal pattern semantic deconstruction to construct a precise semantic feature profile for each abnormal pattern. The multi-objective coordination module is used to obtain preset decision-making objective instructions, perform multi-objective coordination request analysis and dynamic objective priority evaluation on the cognitive entropy change benchmark feature map, and obtain the coordination request priority. The decision boundary module is used to calculate the maximum decision capacity of the cognitive entropy change benchmark feature map, and to calculate the dynamic optimal decision boundary based on the coordination request priority and accurate semantic feature profile, thereby generating a dynamic trade-off constraint range. The constraint solving module is used to solve multi-objective dynamic constraint responses based on the dynamic trade-off constraint range, and to perform real-time decision processing to obtain real-time execution results. The internal control management decision module is used to identify deviations in decision execution response based on real-time execution results, and to perform in-depth optimization of strategy iteration to build an intelligent internal control management decision-making body.

Images