Enterprise research and development auditing management method and system based on artificial intelligence

Abstract

The invention relates to the technical field of database management, in particular to an enterprise research and development auditing management method and system based on artificial intelligence. Comprising the following steps: S1, obtaining enterprise research and development auditing data and an enterprise research and development auditing subject; based on the enterprise research and development auditing main body, extracting an internal auditing main body and an external auditing main body; s2, based on the internal auditing main body, extracting enterprise research and development auditing data of the internal auditing main body and defining the enterprise research and development auditing data as first auditing data; based on the external auditing main body, extracting enterprise research and development auditing data of the external auditing main body and defining the enterprise research and development auditing data as second auditing data; and S3, obtaining intersection data of the first auditing data and the second auditing data based on the first auditing data and the second auditing data. According to the method, AI is used for driving multi-source heterogeneous data integration, hierarchical decoupling and real-time adaptation are adopted, scheduling and format matching are optimized, and integrity and compliance of data streams are ensured.

Description

Technical Field

[0001] The present invention relates to the field of database management technology, and in particular to an enterprise R&D review management method and system based on artificial intelligence. Background Art

[0002] With the rapid growth of multi-source, heterogeneous data, existing database management systems face the following technical bottlenecks: Lack of data stratification: Independent business data and cross-linked data are not decoupled and integrated, resulting in a broken logical chain and an inability to support dynamic temporal correlation analysis; Static rule dependency: Data cleansing and integration rely on preset static rules, making it difficult to dynamically adapt to format conversion requirements across business scenarios; High manual intervention: Internal and external data format conversion relies on manual configuration, resulting in delayed responses and high error rates, and unable to meet real-time compliance verification requirements; Timing conflicts: Traditional time models cannot adaptively adjust to the dynamic data generation cycle and external policy constraints, resulting in inefficient data flow scheduling. These issues make it difficult for existing systems to achieve the technical goals of dynamic adaptation, real-time monitoring, and resource optimization scheduling in the integration of multi-source, heterogeneous data. Summary of the Invention

[0003] In order to overcome the shortcomings of failure in dynamic adaptation of multi-source data, the present invention provides an enterprise R&D review management method and system based on artificial intelligence.

[0004] The technical implementation scheme of the present invention is: an enterprise R&D review management method based on artificial intelligence, comprising the following steps:

[0005] S1: Obtain enterprise R&D audit data and enterprise R&D audit subjects; based on the enterprise R&D audit subjects, extract internal audit subjects and external audit subjects;

[0006] S2: Based on the internal audit subject, extract the enterprise R&D audit data of the internal audit subject and define it as first audit data; based on the external audit subject, extract the enterprise R&D audit data of the external audit subject and define it as second audit data;

[0007] S3: Based on the first audit data and the second audit data, obtain intersection data of the first audit data and the second audit data; based on the intersection data, extract the audit time of the internal audit subject and define it as the first audit time, and simultaneously extract the audit time of the external audit subject and define it as the second audit time;

[0008] S4: Based on the first audit time and the second audit time, obtain the calling relationship of the intersection data; based on the calling relationship, obtain the conversion relationship of the enterprise R&D audit data of the internal audit subject and the external audit subject; based on the conversion relationship, perform process monitoring on the enterprise R&D audit data.

[0009] Preferably, the acquiring of enterprise R&D audit data and enterprise R&D audit subjects; and extracting internal audit subjects and external audit subjects based on the enterprise R&D audit subjects, include:

[0010] Divide enterprise R&D audit data into single-type data and composite-type data;

[0011] Based on the ownership of audit authority, the R&D audit subjects of enterprises are divided into the following two categories: internal audit subjects: departments within the enterprise that exercise audit authority; external audit subjects: organizations that exercise audit authority through external institutions.

[0012] Preferably, based on the internal audit subject, extracting the enterprise R&D audit data of the internal audit subject and defining it as first audit data includes:

[0013] Based on the internal audit subject, the enterprise R&D audit data is sorted in chronological order to obtain a sorting result;

[0014] Marking the sorting result with the single type data and the composite type data to obtain a marking result;

[0015] Based on the marking result, obtaining the time interval of the composite type data and the time interval of the single type data;

[0016] The time interval between the composite type data and the single type data is defined as first audit data.

[0017] Preferably, based on the external audit subject, extracting the enterprise R&D audit data of the external audit subject and defining it as second audit data includes:

[0018] Based on the external audit subject, the enterprise R&D audit data is sorted according to the frequency of format conversion to obtain a sorting result;

[0019] Based on the marking result, obtaining a format conversion frequency ratio of the composite type data and a format conversion frequency ratio of the single type data;

[0020] The format conversion frequency ratio of the composite type data to the single type data is defined as the second audit data.

[0021] Preferably, obtaining intersection data of the first audit data and the second audit data based on the first audit data and the second audit data includes:

[0022] Based on the first audit data and the second audit data, the single type data and the composite type data having the same ratio of the time interval to the format conversion frequency are taken as intersection data.

[0023] Preferably, extracting the audit time of the internal audit subject based on the intersection data and defining it as the first audit time, and extracting the audit time of the external audit subject and defining it as the second audit time, comprises:

[0024] Extracting the time interval between the single type data and the composite type data based on the intersection data;

[0025] Based on the time interval, performing weighted calculation on the maximum time interval mean of the composite type data and the median time interval of the single type data to obtain a dynamic adaptation period;

[0026] generating a first review time based on the dynamic adaptation period and a preset buffer threshold;

[0027] Extracting the format conversion frequency ratio based on the intersection data, and normalizing the maximum frequency ratio of the composite type data to the single type data to obtain a normalized result;

[0028] A second review time is generated based on a mapping relationship between the normalization processing result and the deadline of the external policy document.

[0029] Preferably, obtaining the calling relationship of the intersection data based on the first review time and the second review time includes: obtaining the calling priority of the intersection data according to a calling priority formula, and the calling priority formula is as follows:

[0030]

[0031] Where C is the call priority, T1 is the first review time, T2 is the second review time, and β is the format conversion frequency ratio, which ranges from [0,1].

[0032] Preferably, the conversion relationship between the enterprise R&D audit data of the internal audit subject and the external audit subject is obtained based on the call relationship, including: obtaining the conversion relationship between the enterprise R&D audit data of the internal audit subject and the external audit subject according to a conversion formula, and the conversion formula is as follows:

[0033] F 外 =α·F 内 +η·ΔT

[0034] Among them, F 外 For external audit data format, F 内is the internal audit data format, α is the format conversion coefficient, η is the time correction coefficient, and ΔT is the time calibration coefficient.

[0035] Preferably, the process monitoring of the enterprise R&D audit data based on the conversion relationship includes:

[0036] If the degree of matching between the external audit data format and the policy requirement format is lower than a preset threshold, a format anomaly alert is triggered and the process goes back to the data extraction step;

[0037] If the time calibration coefficient is less than a preset buffer threshold, the dynamic adaptation period is compressed and the calling priority is increased;

[0038] If the conversion efficiency coefficient and the call priority are both lower than the standard, the data will be frozen and a manual review process will be initiated;

[0039] If the complex type data association analysis fails, it is marked as data mismatch and isolated to the exception pool.

[0040] Preferably, an enterprise R&D review management system based on artificial intelligence includes:

[0041] Data acquisition and subject classification module: Real-time acquisition of enterprise R&D audit data, classified into single type and composite type; distinguish between internal and external audit subjects based on audit authority, and mark independent business links and multi-link related data;

[0042] Data extraction and labeling module: sort internal audit data by time and label the type, extract composite data time intervals and single data intervals; sort external audit data by format conversion frequency, and calculate the frequency ratio of composite and single data;

[0043] Data integration and time calculation module: Filters the intersection data of the same interval time and conversion ratio; calculates the dynamic adaptation period based on the time interval, and generates the second audit time based on the normalization of the frequency ratio parameter;

[0044] Dynamic scheduling and monitoring module: Schedules intersection data based on the call priority formula, and dynamically adjusts the internal and external data formats according to the conversion formula; monitors the format matching degree and time calibration coefficient in real time, triggering format anomaly alarms or cycle compression processes.

[0045] Beneficial effects: The present invention integrates the data of internal audit subjects and external audit subjects through a hierarchical decoupling strategy, realizes the classified management of single-type data and composite-type data, and solves the problem of logical chain breaks; extracts composite data time intervals and single data intervals based on time sorting and marking results, generates a dynamic adaptation cycle and superimposes the preset buffer threshold to optimize the generation accuracy of the first audit time; generates a second audit time through format conversion frequency sorting and normalization processing, combined with the external policy document deadline mapping relationship, and improves the efficiency of internal and external data timing coordination; dynamically schedules intersection data based on calling priority formulas, and gives priority to high-timeliness tasks; dynamically adjusts internal and external data formats through conversion formulas to reduce the error rate of manual configuration; monitors format matching and time calibration coefficients in real time, triggers format anomaly alarms or cycle compression processes, accurately isolates mismatched data to the anomaly pool, and ensures the integrity and security of the main database; and ultimately achieves efficient integration, dynamic adaptation and full-link closed-loop monitoring of multi-source heterogeneous data. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flow chart of the enterprise R&D audit management method based on artificial intelligence of the present invention;

[0047] Figure 2 This is a structural diagram of the enterprise R&D audit management system based on artificial intelligence in the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] Example 1: An enterprise R&D review management method based on artificial intelligence, such as Figure 1 As shown, the following steps are included:

[0050] S1: Obtain enterprise R&D audit data and enterprise R&D audit subjects; based on the enterprise R&D audit subjects, extract internal audit subjects and external audit subjects;

[0051] It should be noted that cluster analysis is performed on historical audit data characteristics (such as audit subject type and data format complexity) through machine learning models (such as random forests), internal / external audit subjects are automatically divided, and the classification thresholds of single type and composite type data are optimized to improve the accuracy of data stratification.

[0052] S2: Based on the internal audit subject, extract the enterprise R&D audit data of the internal audit subject and define it as first audit data; based on the external audit subject, extract the enterprise R&D audit data of the external audit subject and define it as second audit data;

[0053] S3: Based on the first audit data and the second audit data, obtain intersection data of the first audit data and the second audit data; based on the intersection data, extract the audit time of the internal audit subject and define it as the first audit time, and simultaneously extract the audit time of the external audit subject and define it as the second audit time;

[0054] It should be noted that the time interval of composite data is dynamically corrected based on the time series prediction algorithm (such as LSTM), and the weight distribution in the weighted calculation is optimized in combination with the historical task completion rate (such as 60%→65%) to generate a more accurate dynamic adaptation cycle.

[0055] S4: Based on the first audit time and the second audit time, obtain the calling relationship of the intersection data; based on the calling relationship, obtain the conversion relationship of the enterprise R&D audit data of the internal audit subject and the external audit subject; based on the conversion relationship, perform process monitoring on the enterprise R&D audit data.

[0056] It should be noted that the reinforcement learning algorithm is used to optimize the β value (format conversion frequency ratio) in the call priority formula, and the parameters are dynamically adjusted according to the real-time task load and policy changes to maximize the resource allocation efficiency; natural language processing (NLP) technology is introduced to parse external policy documents, and the format conversion rule library is automatically updated to ensure that α (format conversion coefficient) matches the policy requirements in real time; the isolation forest algorithm is used to detect composite data correlation anomalies, and the mismatched data is automatically classified and marked (such as "technology-cost conflict") to accelerate the efficiency of the exception pool processing.

[0057] Obtain enterprise R&D audit data and enterprise R&D audit subjects; based on the enterprise R&D audit subjects, extract internal audit subjects and external audit subjects, including:

[0058] Divide enterprise R&D audit data into single-type data and composite-type data;

[0059] Based on the ownership of audit authority, the R&D audit subjects of enterprises are divided into the following two categories: internal audit subjects: departments within the enterprise that exercise audit authority; external audit subjects: organizations that exercise audit authority through external institutions.

[0060] It should be noted that the single type of data refers to data that only involves an independent business link and does not need to be associated with other business links;

[0061] The composite data refers to data that involves multiple business links (K≥2) and needs to be associated with other business links;

[0062] By dividing data types (single / complex) and audit subjects (internal / external), data classification management and clear responsibilities and rights can be achieved, thereby improving audit efficiency and accuracy.

[0063] Data acquisition method: Single-type data: directly extracted from independent business systems (such as financial modules) without cross-system linkage; Composite-type data: integrating multi-link related data through cross-departmental database interfaces (such as R&D and production systems); Audit subject division: internal data is automatically marked by the enterprise OA / ERP system to belong to the department, and external data is obtained through API docking with third-party institutions (such as audit platforms).

[0064] Single data has a simple structure, and independent processing can reduce complexity (such as budget review of a certain department); composite data requires multi-link correlation verification (such as cross-departmental R&D projects that require collaborative review of technology, cost, and progress), and the division facilitates hierarchical decoupling and correlation analysis; the separation of internal and external audit subjects can match different audit rules (such as internal according to corporate processes and external according to policies and regulations) to avoid authority conflicts.

[0065] Examples: Single-type data: independent advertising expense application by the marketing department (involving only budget approval); composite-type data: new product development project (requiring collaborative review by the technical, production, and finance departments); internal audit entity: corporate quality management department; external audit entity: third-party certification body.

[0066] Based on the internal audit subject, the enterprise R&D audit data of the internal audit subject is extracted and defined as first audit data, including:

[0067] Based on the internal audit subject, the enterprise R&D audit data is sorted in chronological order to obtain a sorting result;

[0068] Marking the sorting result with the single type data and the composite type data to obtain a marking result;

[0069] Based on the marking result, obtaining the time interval of the composite type data and the time interval of the single type data;

[0070] The time interval between the composite type data and the single type data is defined as first audit data.

[0071] It should be noted that by time sorting and classification marking, the differences in processing efficiency of internal audit data can be quantified to provide a basis for dynamic scheduling.

[0072] Time sorting: ensure the timing of audit tasks and avoid process confusion; type marking: distinguish data complexity (single / composite) and clarify audit resource allocation rules; interval calculation: identify process bottlenecks and optimize the audit cycle through the time difference between composite type data (multi-link association) and single type data (independent link).

[0073] For example, a scenario involves a company's R&D department submitting two review tasks: Task A (single-type data): an R&D equipment procurement requisition (requiring approval only from the Finance Department); Task B (complex-type data): a new product R&D project (requiring collaborative approval from the Technology, Legal, and Marketing Departments). Operational steps: Sort by time: Task A is submitted on January 1st, and Task B on January 2nd; Mark data types: Task A is marked as "single" and Task B is marked as "complex"; Calculate time intervals: For single-type data (Task A), the Finance Department takes 2 days to complete the review; for complex-type data (Task B), the Technology Department takes 3 days, the Legal Department takes 2 days, and the Marketing Department takes 2 days, for a total of 7 days; Define the first review data: The intervals for single-type data (2 days) and complex-type data (7 days) are recorded as baseline parameters for subsequent dynamic scheduling. Correlation analysis: If the interval between composite data types is significantly higher than that between single data types (e.g., 7 days vs. 2 days), resource allocation will be automatically triggered (e.g., increasing the number of concurrent reviewers for composite tasks). In subsequent tasks, if new composite tasks are submitted, the "7-day" benchmark will be used as a reference to prioritize the allocation of review resources to shorten the cycle.

[0074] Based on the external audit subject, the enterprise R&D audit data of the external audit subject is extracted and defined as second audit data, including:

[0075] Based on the external audit subject, the enterprise R&D audit data is sorted according to the frequency of format conversion to obtain a sorting result;

[0076] Based on the marking result, obtaining a format conversion frequency ratio of the composite type data and a format conversion frequency ratio of the single type data;

[0077] The format conversion frequency ratio of the composite type data to the single type data is defined as the second audit data.

[0078] It should be noted that by sorting the frequency of format conversion and calculating the ratio, the format adaptation cost of external audit data can be quantified, and the conversion rules can be optimized to reduce manual intervention.

[0079] Frequency sorting: Identify data types with high frequency conversion and prioritize adapting to their format rules; Ratio calculation: Composite type data involves multiple links and needs to adapt to different format requirements, resulting in its conversion frequency usually being higher than that of single type data; Define second audit data: Reflect the complexity of format conversion through ratios, which are used to dynamically adjust resource allocation for the external audit process.

[0080] Example scenario: A company submits two types of review tasks to an external organization: Task C (single-type data): R&D equipment procurement contract (only needs to be converted to PDF and submitted to the Industry and Commerce Bureau); Task D (complex-type data): New product launch application (needs to be converted to PDF, XML, and JSON formats, and submitted to the Food and Drug Administration, the Patent Office, and the industry association respectively). Operation steps: Sort by conversion frequency: Task D requires conversion to three formats (PDF, XML, and JSON), with a frequency of three times per month; Task C only requires conversion to one format (PDF), with a frequency of once per month;

[0081] Calculate the frequency ratio: The format conversion frequency ratio of composite data (Task D) to single data (Task C) is 3:1. Define the second audit data: This ratio "3:1" is recorded as a key parameter for external audits and used for subsequent process optimization. Correlation analysis: If the format conversion frequency ratio of composite data is significantly high (e.g., 3:1), automatically generate multi-format templates to reduce manual repetitive operations. In subsequent tasks, if similar composite data is submitted, pre-allocate conversion resources based on the "3:1" ratio (e.g., call an API to batch generate XML / JSON), shortening the external audit cycle.

[0082] Obtaining intersection data of the first audit data and the second audit data based on the first audit data and the second audit data includes:

[0083] Based on the first audit data and the second audit data, the single type data and the composite type data having the same ratio of the time interval to the format conversion frequency are taken as intersection data.

[0084] It should be noted that through cross-dimensional feature matching (time interval + format conversion ratio), data with consistent internal and external audit characteristics can be screened to achieve accurate resource allocation and process standardization.

[0085] Intersection definition: The data must meet the requirements of the same internal audit time and the same external format conversion frequency ratio; classification integration: grouping similar feature data to facilitate batch processing or priority scheduling; optimization goal: reduce duplicate calculations and improve the collaborative efficiency across internal and external audits.

[0086] For example, a scenario involves an enterprise's historical audit data, including the following: Task A (single-type data): internal audit time of 2 days, external format conversion frequency of 1 / month (ratio of 1:1); Task B (complex-type data): internal audit time of 7 days, external format conversion frequency of 3 / month (ratio of 3:1); Task E (single-type data): internal audit time of 2 days, external format conversion frequency of 1 / month (ratio of 1:1); Task F (complex-type data): internal audit time of 7 days, external format conversion frequency of 3 / month (ratio of 3:1). Operation steps: Match the intersection data: Task A and Task E (internal time of 2 days + external format conversion ratio of 1:1) → single-type intersection; Task B and Task F (internal time of 7 days + external format conversion ratio of 3:1) → composite-type intersection. Application scenario: Single-type intersection: automatically assigned to a fast track (such as a templated audit process) and completed within 2 days; composite-type intersection: triggering a cross-departmental collaborative audit mechanism, pre-allocating multi-format conversion resources, and shortening the cycle to 5 days (originally 7 days). Correlation analysis: Intersection data reveals common patterns in audit efficiency. For example, composite data with a "ratio of 3:1 + time of 7 days" requires strengthened cross-departmental collaboration. Subsequent similar tasks directly call on the optimization strategy of historical intersection data to reduce manual configuration costs.

[0087] The extracting, based on the intersection data, the audit time of the internal audit subject and defining it as the first audit time, and the extracting, based on the intersection data, the audit time of the external audit subject and defining it as the second audit time, includes:

[0088] Extracting the time interval between the single type data and the composite type data based on the intersection data;

[0089] Based on the time interval, performing weighted calculation on the maximum time interval mean of the composite type data and the median time interval of the single type data to obtain a dynamic adaptation period;

[0090] generating a first review time based on the dynamic adaptation period and a preset buffer threshold;

[0091] Extracting the format conversion frequency ratio based on the intersection data, and normalizing the maximum frequency ratio of the composite type data to the single type data to obtain a normalized result;

[0092] A second review time is generated based on a mapping relationship between the normalization processing result and the deadline of the external policy document.

[0093] It's important to note that the calculation of the dynamic adaptive period can essentially be abstracted as a vector dot product operation: a feature vector is constructed by combining the mean of the maximum time intervals of the composite data type with the median of the time intervals of the single data type. This feature vector is then dot-producted with a preset weight vector. The resulting scalar result is then added to the preset buffer threshold to generate the first review time. This operation transforms the weighted calculation of time intervals into an efficient mapping in vector space, ensuring the mathematical rigor and dynamic adaptability of period generation while maintaining logical consistency with the hierarchical decoupling strategy.

[0094] Through the fusion of multi-dimensional parameters, the efficiency characteristics (time, format) of internal and external audit data are dynamically bound to external policy constraints to generate an accurate audit schedule.

[0095] Parameter acquisition method: Maximum time interval mean: take the maximum value from the historical review cycle of composite type data and calculate the mean (such as the 7-day period mean of tasks B and F); time interval median: take the median from the time series of single type data (such as the 2-day median of tasks A and E); maximum format conversion frequency of composite type data: the highest value of the number of format conversions in the composite data (such as 3 times / month for task D); maximum format conversion frequency of single type data: the highest value of the number of format conversions in the single data (such as 1 time / month for task C); external policy deadline: obtained through policy document parsing or API interface (such as the Drug Administration requires that the deadline for new drug marketing review is June 30).

[0096] Parameter setting method: Preset buffer threshold: set according to business fault tolerance requirements (such as reserving a 20% time buffer); dynamic adaptation period: weighted formula (mean of the maximum time interval of composite type data × 60% + median of the time interval of single type data × 40%); mapping relationship: linearly map the normalized result to the time window of the policy deadline (the normalized result is linearly mapped to the time window of the policy deadline, for example, a normalized value of 0.8 corresponds to 10 days before the deadline).

[0097] Example, scenario: Generate audit time based on intersection data (composite data of tasks B and F, single type data of tasks A and E): Extract time parameters: mean of maximum time interval of composite type data: 7 days (mean of tasks B and F); median of time interval of single type data: 2 days (median of tasks A and E); Calculate dynamic adaptation cycle: weighted calculation: 7 days × 60% + 2 days × 40% = 5 days; superimpose preset buffer threshold (20%): 5 days × 1.2 = 6 days → first audit time; Extract format parameters: the maximum format conversion frequency of composite type data is 3 times / month (task D); the maximum format conversion frequency of single type data is 1 time / month (task C); Normalization processing: based on the frequency range of composite and single types (1-3 times / month

[0098] Monthly), map Task D's 3 times / month to 1.0; this mapping generates the second review time: the policy deadline is June 30th, and the normalized value 1.0 is mapped to 5 days before the deadline → the second review time is June 25th. Correlation application: If a new composite task is submitted (such as Task G, which requires three format conversions), directly call the dynamic adaptation cycle of 6 days and map it to 5 days before the policy deadline, without repeated calculation. Based on the mapping rule between the normalization result and the policy deadline (e.g., 5 days before the deadline), the second review time is automatically adjusted to July 5th.

[0099] Based on the first review time and the second review time, obtaining the calling relationship of the intersection data includes: obtaining the calling priority of the intersection data according to a calling priority formula, the calling priority formula is as follows:

[0100]

[0101] Where C is the call priority, T1 is the first review time, T2 is the second review time, and β is the format conversion frequency ratio, which ranges from [0,1].

[0102] It should be noted that the call priority formula dynamically allocates audit resources by quantifying the correlation between time urgency and data conversion requirements.

[0103] When T1 ≤ T2: Priority is positively correlated with conversion frequency (β) and internal review time (T1), and negatively correlated with time difference (T2-T1). The tighter the time window (the smaller T2-T1) or the higher the conversion demand (the larger β), the higher the priority. When T1 > T2: Internal review times out, priority is reset to zero to avoid wasted resources.

[0104] Example, scenario: Parameters based on Task B (complex type data) and Task C (single type data): Task B (new product development): T1 = 7 days (first review time), T2 = 10 days (second review time mapped to the policy deadline); β = 0.8 (format conversion frequency ratio is 3:1, normalized to 0.8); priority calculation: Result: High priority, triggering cross-departmental collaborative review. Task C (equipment procurement): T1 = 2 days, T2 = 5 days; β = 0.2 (format conversion frequency ratio is 1:1, normalized to 0.2); Priority calculation: Result: Low priority, assigned to the standardization review queue. Related application: If Task B's T1 is extended to 12 days (T2 = 10 days), then Priority C = 0, automatically freezing the task and issuing an alert. If Task C's β increases to 0.6 (e.g., due to new format requirements), the priority increases to 0.4, triggering resource pre-allocation.

[0105] Based on the call relationship, the conversion relationship between the enterprise R&D audit data of the internal audit subject and the external audit subject is obtained, including: obtaining the conversion relationship between the enterprise R&D audit data of the internal audit subject and the external audit subject according to a conversion formula, the conversion formula is as follows:

[0106] F 外 =α·F 内 +η·ΔT

[0107] Among them, F 外 For external audit data format, F 内 is the internal audit data format, α is the format conversion coefficient, η is the time correction coefficient, and ΔT is the time calibration coefficient.

[0108] It should be noted that the conversion formula achieves dynamic adaptation of internal and external audit data by integrating format rules and time sensitivity.

[0109] Parameter Definitions: α (Format Conversion Factor): The weight of the rules for converting the internal format to the external format (e.g., field mapping ratio); η (Time Correction Factor): The weight of the impact of time urgency on format conversion (e.g., the need to simplify the format due to an approaching deadline); ΔT (Time Alignment Factor): The difference in internal and external review times (ΔT = T2 - T1), reflecting time window pressure. Formula Logic: The external format is converted from the internal format according to the rules (α term), and adjusted based on time pressure (η·ΔT term) to ensure a balance between format compliance and timeliness.

[0110] Example, scenario: Task B (new product development project) needs to convert internal data into the XML format required by the Food and Drug Administration: Parameter assignment: F 内 : Internal JSON format (including technical parameters, cost, and schedule); α = 0.9: JSON → XML field mapping covers 90% of the content; η = 0.1: For every additional day of time pressure, 10% of non-critical fields are simplified; ΔT = T2 - T1 = 10 days - 7 days = 3 days (time difference). Conversion calculation: F 外 =0.9*F 内 +0.1*3=0.9F 内 +0.3, result: 90% of internal fields are fully converted (such as technical parameters, costs); because ΔT = 3, 10% × 3 = 30% of non-critical fields are simplified (such as deleting progress details and only retaining milestones). Output format: Drug Administration XML only contains core parameters and milestones, meeting timeliness requirements. Correlation application: If ΔT = 5 (time is more urgent), the formula output is F 外 =0.9F 内 +0.5, further simplifying 50% of non-critical fields; if policy requirements change (α drops to 0.7), automatically adjust mapping rules to ensure format compliance.

[0111] Based on the conversion relationship, process monitoring is performed on the enterprise R&D audit data, including:

[0112] If the degree of matching between the external audit data format and the policy requirement format is lower than a preset threshold, a format anomaly alert is triggered and the process goes back to the data extraction step;

[0113] If the time calibration coefficient is less than a preset buffer threshold, the dynamic adaptation period is compressed and the calling priority is increased;

[0114] If the conversion efficiency coefficient and the call priority are both lower than the standard, the data will be frozen and a manual review process will be initiated;

[0115] If the complex type data association analysis fails, it is marked as data mismatch and isolated to the exception pool.

[0116] It should be noted that the exception pool setting method includes data mismatch marking: data that fails correlation analysis is automatically isolated to an independent storage area; manual review triggering: after freezing the data, it is pushed to the designated management interface and a responsible person is assigned.

[0117] Through multi-condition threshold monitoring and automated process control, abnormal data can be quickly isolated and resources reallocated to ensure the integrity and compliance of the audit chain.

[0118] For example, in the scenario of the review process for Task D (New Product Launch Application), the following steps are used: Format Anomaly Alert: If the external data format match reaches 70% (threshold 80%), an alert is triggered and the data extraction module is backtracked to regenerate the XML file. Time Pressure Response: The time calibration coefficient ΔT = 1 day (preset buffer threshold 2 days), compressing the dynamic adaptation cycle from 7 days to 5 days, and increasing the priority from 1.87 to 2.5. Freezing and Manual Review: With a conversion efficiency coefficient of 0.4 (standard 0.5) and a priority of 0.2 (standard 0.3), data is frozen and pushed to the Quality Department for manual review. Data Mismatch Isolation: If complex data correlation analysis fails (e.g., conflicts between technical parameters and cost data), it is marked as a mismatch and isolated to the exception pool, pending correction and resubmission by the Technical Department. Correlation Application: If the format match reaches 75% for Task B (R&D Project), the backtracking process is also triggered. When data backlogs in the exception pool, the responsible person is automatically notified to address them, avoiding process bottlenecks.

[0119] Example 2: Based on Example 1, an enterprise R&D review management system based on artificial intelligence, such as Figure 2 Shown, including:

[0120] Data acquisition and subject classification module: Real-time acquisition of enterprise R&D audit data, classified into single type and composite type; distinguish between internal and external audit subjects based on audit authority, and mark independent business links and multi-link related data;

[0121] Data extraction and labeling module: sort internal audit data by time and label the type, extract composite data time intervals and single data intervals; sort external audit data by format conversion frequency, and calculate the frequency ratio of composite and single data;

[0122] Data integration and time calculation module: Filters the intersection data of the same interval time and conversion ratio; calculates the dynamic adaptation period based on the time interval, and generates the second audit time based on the normalization of the frequency ratio parameter;

[0123] Dynamic scheduling and monitoring module: Schedules intersection data based on the call priority formula, and dynamically adjusts the internal and external data formats according to the conversion formula; monitors the format matching degree and time calibration coefficient in real time, triggering format anomaly alarms or cycle compression processes.

[0124] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An enterprise R&D review management method based on artificial intelligence, characterized by: The following steps are involved: S1: Obtain enterprise R&D audit data and enterprise R&D audit subjects; based on the enterprise R&D audit subjects, extract internal audit subjects and external audit subjects; S2: Based on the internal audit subject, extract the enterprise R&D audit data of the internal audit subject and define it as first audit data; based on the external audit subject, extract the enterprise R&D audit data of the external audit subject and define it as second audit data; S3: Based on the first audit data and the second audit data, obtain intersection data of the first audit data and the second audit data; based on the intersection data, extract the audit time of the internal audit subject and define it as the first audit time, and simultaneously extract the audit time of the external audit subject and define it as the second audit time; S4: Based on the first review time and the second review time, obtaining a call relationship of the intersection data; Based on the call relationship, a conversion relationship between the enterprise R&D audit data of the internal audit subject and the external audit subject is obtained; Based on the conversion relationship, the process of the enterprise R&D audit data is monitored.

2. The enterprise R&D review management method based on artificial intelligence as claimed in claim 1 is characterized by: The acquisition of enterprise R&D audit data and enterprise R&D audit subjects; Based on the enterprise R&D audit subject, the internal audit subject and the external audit subject are extracted, including: Divide enterprise R&D audit data into single-type data and composite-type data; Based on the ownership of audit authority, the R&D audit subjects of enterprises are divided into the following two categories: internal audit subjects: departments within the enterprise that exercise audit authority; external audit subjects: organizations that exercise audit authority through external institutions.

3. The enterprise R&D review management method based on artificial intelligence as claimed in claim 2 is characterized by: The extracting, based on the internal audit subject, the enterprise R&D audit data of the internal audit subject and defining it as first audit data includes: Based on the internal audit subject, the enterprise R&D audit data is sorted in chronological order to obtain a sorting result; Marking the sorting result with the single type data and the composite type data to obtain a marking result; Based on the marking result, obtaining the time interval of the composite type data and the time interval of the single type data; The time interval between the composite type data and the single type data is defined as first audit data.

4. The enterprise R&D review management method based on artificial intelligence as claimed in claim 3 is characterized by: The extracting, based on the external audit subject, the enterprise R&D audit data of the external audit subject and defining it as second audit data includes: Based on the external audit subject, the enterprise R&D audit data is sorted according to the frequency of format conversion to obtain a sorting result; Based on the marking result, obtaining a format conversion frequency ratio of the composite type data and a format conversion frequency ratio of the single type data; The format conversion frequency ratio of the composite type data to the single type data is defined as the second audit data.

5. The enterprise R&D review management method based on artificial intelligence as claimed in claim 4 is characterized in that: The obtaining, based on the first audit data and the second audit data, intersection data of the first audit data and the second audit data, includes: Based on the first audit data and the second audit data, the single type data and the composite type data having the same ratio of the time interval to the format conversion frequency are taken as intersection data.

6. The enterprise R&D review management method based on artificial intelligence as claimed in claim 3 is characterized by: The extracting, based on the intersection data, the audit time of the internal audit subject and defining it as the first audit time, and the extracting, based on the intersection data, the audit time of the external audit subject and defining it as the second audit time, includes: Based on the intersection data, extracting the time interval between the single type data and the composite type data; based on the time interval, performing a weighted calculation on the maximum time interval mean of the composite type data and the median time interval of the single type data to obtain a dynamic adaptation period; generating a first review time based on the dynamic adaptation period and a preset buffer threshold; Extracting the format conversion frequency ratio based on the intersection data, and normalizing the maximum frequency ratio of the composite type data to the single type data to obtain a normalized result; A second review time is generated based on a mapping relationship between the normalization processing result and the deadline of the external policy document.

7. The enterprise R&D review management method based on artificial intelligence as claimed in claim 1 is characterized by: The obtaining of the calling relationship of the intersection data based on the first review time and the second review time includes: obtaining the calling priority of the intersection data according to a calling priority formula, the calling priority formula being as follows: Where C is the call priority, T1 is the first review time, T2 is the second review time, and β is the format conversion frequency ratio, which ranges from [0,1].

8. The enterprise R&D review management method based on artificial intelligence as claimed in claim 1 is characterized by: The conversion relationship between the enterprise R&D audit data of the internal audit subject and the external audit subject is obtained based on the call relationship, including: obtaining the conversion relationship between the enterprise R&D audit data of the internal audit subject and the external audit subject according to a conversion formula, the conversion formula is as follows: F 外 =α·F 内 +η·ΔT Among them, F 外 For external audit data format, F 内 is the internal audit data format, α is the format conversion coefficient, η is the time correction coefficient, and ΔT is the time calibration coefficient.

9. The enterprise R&D review management method based on artificial intelligence as claimed in claim 8 is characterized by: The process monitoring of the enterprise R&D audit data based on the conversion relationship includes: If the degree of matching between the external audit data format and the policy requirement format is lower than a preset threshold, a format anomaly alert is triggered and the process goes back to the data extraction step; If the time calibration coefficient is less than a preset buffer threshold, the dynamic adaptation period is compressed and the calling priority is increased; If the conversion efficiency coefficient and the call priority are both lower than the standard, the data will be frozen and a manual review process will be initiated; If the complex type data association analysis fails, it is marked as data mismatch and isolated to the exception pool.

10. An enterprise R&D review management system based on artificial intelligence, used to implement the enterprise R&D review management method based on artificial intelligence as described in any one of claims 1 to 9, characterized in that: include: Data acquisition and subject classification module: real-time acquisition of enterprise R&D review data, classified by single type and composite type; Distinguish internal and external audit entities based on audit authority, and mark independent business links and multi-link related data; Data extraction and labeling module: sort internal audit data by time and label the type, extract composite data time intervals and single data intervals; sort external audit data by format conversion frequency, and calculate the frequency ratio of composite and single data; Data integration and time calculation module: Filters the intersection data of the same interval time and conversion ratio; calculates the dynamic adaptation period based on the time interval, and generates the second audit time based on the normalization of the frequency ratio parameter; Dynamic scheduling and monitoring module: Schedules intersection data based on the call priority formula, and dynamically adjusts the internal and external data formats according to the conversion formula; monitors the format matching degree and time calibration coefficient in real time, triggering format anomaly alarms or cycle compression processes.

Images