Multi-angle intelligent calculation and exception collection risk early warning method and system for research and development expenses

Abstract

The present application relates to the technical field of data processing, in particular to a multi-angle intelligent calculation and abnormal collection risk early warning method and system for research and development expenses. The present application automatically associates each cost expenditure to the corresponding research and development project through multi-source data collection and standardized cleaning, constructs a basic database with unified data structure, and solves the problem that multi-source heterogeneous data cannot be automatically recognized and attributed by the system due to inconsistent field specifications and lack of project association identifier. Differentiated allocation calculation models are called for personnel manual, equipment depreciation, power consumption, direct materials, intangible asset amortization and other expenses. The power consumption scenario is compatible with instrument reading direct measurement and unit energy consumption coefficient conversion of two data formats, and supports single device multi-energy independent accounting, solves the problem of incompatible multi-type energy consumption data formats and single model cannot adapt to all data formats, and enables different formats of energy consumption data to complete allocation calculation under a unified framework.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a method and system for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses. Background Technology

[0002] In the field of R&D expense accounting and management, enterprises need to collect and account for R&D expenditures according to six major categories: personnel costs, direct input costs, fixed asset depreciation costs, intangible asset amortization costs, new product design and testing fees, and other related expenses. Currently, the mainstream accounting method in the industry still adopts a model of manual ledgers in Excel combined with manual summarization of time allocation tables. That is, finance personnel and R&D personnel fill in the forms offline and manually allocate the R&D expenses of each project to complete the collection.

[0003] This traditional accounting model has the following technical problems:

[0004] Multi-source heterogeneous R&D data is scattered across multiple business systems, with inconsistent field specifications and coding rules across systems. Furthermore, the raw data generally lacks R&D project association identifiers, preventing computer systems from automatically identifying cost allocation and requiring manual cross-system processing before entering the calculation process. Costs of personnel and equipment shared across projects rely on manual breakdown based on working hours, resulting in non-unique calculation paths and a lack of automatic verification, making errors prone to occur and difficult to locate automatically. Energy consumption data presents various incompatible formats, such as instrument readings, unit energy consumption coefficients, and parallel processing of multiple energy sources. Existing systems lack a unified calculation model compatible with all these formats, leading to allocation results deviating from actual consumption. During manual accounting, issues such as conflicting working hour logic and lack of basis for cost allocation only surface after tax audits or audits, with the system lacking real-time quality verification throughout data processing. More importantly, the final output cost data must simultaneously meet four sets of regulatory requirements: accounting, high-tech enterprise certification, additional deductions, and R&D (research and development) statistics. Each set of structures differs in the definition, scope, and limits of cost items, leading to frequent issues of mixed definitions and inconsistent data when manually compiled.

[0005] To address the aforementioned issues, this invention proposes a method and system for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses. Summary of the Invention

[0006] To address the aforementioned problems in the existing technology, the purpose of this invention is to provide a method and system for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses.

[0007] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0008] A multi-caliber intelligent calculation and anomaly collection risk warning method for R&D expenses includes:

[0009] Collect raw R&D data from various business systems of the enterprise, perform standardized cleaning and project matching audit on the raw R&D data, establish a link between each cost expenditure and the corresponding R&D project, and lock it into a standardized accounting basic database;

[0010] Based on the standardized accounting database, differentiated allocation calculation models are called according to different expense categories to collect the R&D expenses of each project, and the collected expense data of each category is obtained.

[0011] The collected expense data of each subject is filled into the corresponding standard auxiliary ledger table, and the R&D expenditure summary and R&D expense discount details are generated level by level. Based on four different mapping rules, it is simultaneously converted into four output structures: accounting caliber, high-tech enterprise certification caliber, additional deduction caliber and R&D statistical caliber, generating a summary table of expenses for each project and a detailed ledger of expenses for each project.

[0012] During the data acquisition and allocation calculation process, a two-layer anomaly check is performed. The first layer of check acts as the entry gate for the four sets of mapping rules, verifying the credibility of the data source entering the mapping link. Data that fails the first layer of check is prevented from flowing into the mapping link. The second layer of check acts as the exit filter for the mapping link, verifying the structural rationality of the data output by the mapping link.

[0013] When an anomaly is triggered, a correction and recalculation process is executed. The corrected data is re-entered into the mapping link for recalculation. Based on historical early warning data, the mapping parameters of each expense item under each caliber are optimized. After confirming that all data for the month is correct, monthly data locking is executed, forming a complete audit chain.

[0014] Preferably, the standard auxiliary ledgers include at least the raw and auxiliary material expense statement, the fuel and power consumption summary statement, the equipment depreciation expense summary statement, the intangible asset amortization expense summary statement, and other expense summary statements; the four different mapping rules include at least the following: for the same raw and auxiliary material expense, the additional deduction is recorded in full based on the actual consumption, while the accounting method is recorded based on the quantity used; for outsourced R&D expenses, the additional deduction is only allowed to be recorded at 80% of the actual payment amount, while other methods are recorded in full based on the actual payment amount.

[0015] Preferably, for other expense items, the additional deduction caliber is solved using algebraic methods to address circular dependencies:

[0016] The five expenses of all R&D projects are summed up. These five expenses include personnel costs, direct input costs, depreciation costs, amortization of intangible assets, and new product design fees. According to the additional deduction policy, there is a circular dependency relationship between the other expense limit and the sum of the five expenses. The circular dependency relationship is transformed into a form that can be solved in one go through algebraic transformation to calculate the other expense limit for all projects. The actual total amount of other expenses incurred by all projects is compared with the limit. When the actual total amount incurred exceeds the limit, the limit is allocated to each project according to the proportion of the actual amount of other expenses incurred by each project to the total amount of other expenses incurred by all projects, and a consistency check is performed.

[0017] Preferably, the first layer of verification is a logical consistency verification, which is initiated during the data acquisition phase. It identifies logical conflicts such as R&D hours exceeding the total monthly attendance hours of employees and equipment R&D hours exceeding the total monthly operating time of equipment, and triggers an overtime warning. The second layer of verification is a cost structure verification, which is initiated after the allocation calculation and four sets of mappings are completed. It compares the cost item ratio of each project with the preset reasonable ratio threshold range. When the ratio of a certain item exceeds the threshold range, a cost structure warning is triggered.

[0018] Preferably, when the output data of a certain caliber triggers a cost structure warning, if only a single caliber has a structural anomaly, the output of that caliber is marked as requiring manual review, while other calibers are output normally; if multiple calibers have structural anomalies at the same time, a correction and recalculation process is triggered.

[0019] Preferably, the mapping parameters of each expense item under each scope are optimized, specifically as follows:

[0020] The mapping rules are abstracted into parameterized functions, with the optimization objective of minimizing the weighted sum of squared deviations between the mapping ratio of each historical period and the boundary of the compliance threshold range, and the optimal inclusion ratio of the item under the corresponding caliber is automatically calculated.

[0021] The deviation value for each historical period is determined as follows:

[0022] When the mapping ratio of a period is within the compliant threshold range, the deviation is zero. When it is below the lower limit, the deviation is the lower limit minus the mapping ratio. When it is above the upper limit, the deviation is the mapping ratio minus the upper limit. The weight of each period is determined using an exponential decay strategy, with recent data having a higher weight than older data.

[0023] Preferably, monthly data locking is performed by organizational unit and year / month. Once locked, modifications to all measurement and summary data for that month are controlled. The entire audit chain consists of modification logs from each of the aforementioned stages.

[0024] Preferably, it also includes a visual display, including:

[0025] The completed cost data will be displayed in the form of a cost structure ring diagram, monthly cost curves, and project status distribution map; it also includes a data loop of R&D input and output: from the start of the R&D project, through the association of project, personnel and equipment, as well as process measurement, monthly summary and locking, standard reports and application output, to the R&D results and results transformation records, all data are connected step by step along the entire life cycle of the project.

[0026] This invention also discloses a multi-caliber intelligent settlement and anomaly collection risk early warning system for R&D expenses, including:

[0027] The data collection and cleaning module is configured to collect raw R&D data from various business systems of the enterprise, perform standardized cleaning and project matching audit on the raw R&D data, establish a link between each cost expenditure and the corresponding R&D project, and lock it into a standardized accounting basic database.

[0028] The settlement and mapping module is configured to, based on the standardized accounting database, call different allocation calculation models according to different expense items to complete the collection of R&D expenses for each project, fill the collected expense data of each item into the corresponding standard auxiliary ledger, and generate a summary of R&D expenditures and details of R&D expense discounts by level. According to four different mapping rules, it is simultaneously converted into four output structures: accounting caliber, high-tech enterprise certification caliber, additional deduction caliber, and R&D statistical caliber, generating a summary table of expenses for each project and a detailed ledger of expenses for each project.

[0029] The verification and early warning module is configured to run a two-layer anomaly verification during the data acquisition and allocation calculation process. The first layer of verification acts as the entry gate for the four sets of mapping rules, verifying the credibility of the data source entering the mapping link. Data that fails the first layer of verification is prevented from flowing into the mapping link. The second layer of verification acts as the exit filter for the mapping link, verifying the structural rationality of the data output by the mapping link.

[0030] The closed-loop and optimization module is configured to execute a correction and recalculation process when an anomaly is triggered. The corrected data is re-entered into the mapping link for recalculation, and the mapping parameters of each expense item under each caliber are optimized based on historical early warning data.

[0031] The version finalization and auditing module is configured to lock the monthly data after confirming that all data for the month is correct, thus forming a complete audit chain.

[0032] Preferably, the closed-loop and optimization module optimizes the mapping parameters of each expense item under each caliber. Specifically, the mapping rules are abstracted into parameterized functions, with the optimization objective being to minimize the sum of squared weighted deviations between the mapping ratio of each historical period and the boundary of the compliance threshold interval. The optimal inclusion ratio of the item under the corresponding caliber is then automatically calculated.

[0033] The deviation value for each historical period is determined as follows: when the mapping ratio of the period is within the compliance threshold range, the deviation is zero; when it is below the lower limit, the deviation is the lower limit minus the mapping ratio; when it is above the upper limit, the deviation is the mapping ratio minus the upper limit. The weight of each period is determined using an exponential decay strategy, with recent data having a higher weight than older data.

[0034] Beneficial effects

[0035] 1. This invention automatically links each cost expenditure to the corresponding R&D project through multi-source data acquisition and standardized cleaning, constructing a basic database with a unified data structure. This solves the problem that multi-source heterogeneous data cannot be automatically identified and attributed by the system due to inconsistent field specifications and lack of project association identifiers. On this basis, differentiated allocation calculation models are called for personnel labor, equipment depreciation, power consumption, direct materials, intangible asset amortization and other expenses. Among them, the power consumption scenario is compatible with two data formats: direct meter reading and unit energy consumption coefficient conversion. It also supports independent accounting of multiple energy sources for a single device, solving the problem of incompatibility of multiple types of energy consumption data formats and the inability of a single model to adapt to all data formats. This enables energy consumption data of different formats to be allocated and calculated under a unified framework.

[0036] 2. This invention constructs a data quality tiered assurance mechanism through a quality cascading constraint relationship between a two-layer anomaly check and a four-caliber parallel mapping link. Specifically: the first-layer check is initiated during the data acquisition phase, performing a logical consistency comparison on raw data such as work hours and attendance. Data that fails the check is marked as unreliable and prevented from flowing into the mapping link, thus avoiding systemic risks caused by data distortion at the source. Data that passes the first-layer check is allowed to enter the mapping link for allocation and multi-caliber conversion. The second-layer check is initiated at the output end of the mapping link, automatically comparing the subject proportions of each caliber's output data with preset threshold ranges. When a single caliber is anomaly, only that caliber is marked as pending review without blocking the other three outputs. This mechanism ensures that data entering the mapping link has logical credibility at the source, and that the output multi-caliber data has structural rationality at the result end. It solves the problem in the data processing flow of being unable to distinguish the credibility of the data source and the rationality of the calculation results due to the lack of layered checks, resulting in a significantly higher early warning accuracy than single-layer check schemes.

[0037] 3. This invention summarizes data level by level along expense categories, standard auxiliary ledgers, R&D expenditure summaries, and R&D expense discount details. Based on four different sets of field mapping rules, it simultaneously converts the same set of standardized data into four output structures: accounting, high-tech enterprise certification, additional deduction, and R&D statistics. Field differences between structures can be directly traced back to the corresponding mapping rule level, solving the problem of rule inconsistencies and data inconsistencies caused by manual compilation when outputting the same set of basic data to multiple regulatory standards. At the same time, monthly data locking switches the status of finalized data from editable to locked. After locking, data modification is controlled, and combined with modification logs, a full change traceability path is formed from the original collection to the final version, ensuring that the final output multi-standard data is guaranteed in both rule consistency and change traceability. Attached Figure Description

[0038] Figure 1 This is a flowchart of the method provided by the present invention;

[0039] Figure 2 This is a system module diagram provided by the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.

[0041] Example 1

[0042] Please see Figure 1 As shown in the figure, this embodiment provides a method for intelligent settlement and anomaly collection risk warning of R&D expenses with multiple calibers. The method includes the following steps:

[0043] S1, Comprehensive and Multi-Source R&D Data Acquisition;

[0044] Collect all raw data related to R&D activities from all existing business systems within the enterprise. The specific implementation method is as follows:

[0045] The API interface automatically connects to OA system, ERP system, R&D management system and equipment monitoring system, and retrieves personnel attendance and salary information, equipment ledger and operation records, raw and auxiliary material requisition and waste recycling records, fuel and power consumption data, intangible asset amortization information and outsourcing contract execution plans from the above systems.

[0046] Preferably, the personnel attendance record includes the start and end times and working hours of each personnel in each R&D project, and has a confirmation status field; the salary record includes details of salary and social security and housing provident fund contributions; the waste recycling requisition record includes the type, quantity and unit price of recyclable materials.

[0047] For offline paper documents not integrated into the business system, data is collected manually. Furthermore, both attendance records and requisition records support both single-entry and batch import methods. Optionally, this method also collects rest day configuration information, including statutory holidays and compensatory leave calendars, to eliminate the impact of statutory rest days on work hour data in subsequent work hour calculations.

[0048] After completing the above data collection, a raw data set covering the entire R&D process is obtained. This data set will be used as the processing object for subsequent steps in the data cleaning and standardization process.

[0049] S2. Data standardization and cleaning, and project matching review;

[0050] The collected raw data is sent to a data standardization and cleaning process to ensure that the multi-source heterogeneous data has a unified format and project attribution identifier before entering the allocation calculation. Specifically, based on the project number, R&D cycle, and R&D scope contained in the R&D project initiation report as the matching benchmark, operating expenses unrelated to R&D activities (including but not limited to production expenses, administrative expenses, and sales expenses) and expense records occurring outside the R&D cycle are removed. Then, the field format, unit of measurement, and statistical caliber of each data source are standardized, and the missing project attribution identifier is filled in according to the project binding information in the attendance record, thus establishing a link between each cost expenditure and the corresponding R&D project.

[0051] After automatic cleaning and matching are completed, the results are pushed to the manual review stage. Preferably, the automatic matching results are jointly verified by R&D personnel and finance personnel. Once the review is passed, the results are locked into the standardized accounting foundation database. This database is the unified data source for all subsequent allocation calculations and caliber outputs, and its completeness and accuracy directly determine the processing quality of subsequent steps in the entire method.

[0052] S3, multimodal computation engine and four-caliber parallel mapping output;

[0053] Based on the standardized accounting database, differentiated calculation models are invoked according to different expense categories to collect the R&D expenses of each project. Specifically:

[0054] For personnel costs and equipment depreciation, the monthly total salary and monthly equipment depreciation are allocated to the corresponding projects based on the proportion of man-hours invested by personnel or equipment in each R&D project. Personnel costs include wages and social security / housing provident fund contributions. This approach achieves accurate aggregation of personnel and equipment costs based on actual R&D man-hours invested.

[0055] For power energy consumption costs, the system is compatible with processing two data formats: for equipment with independent metering instruments, the instrument readings are used for direct measurement; for equipment that only counts the unit's comprehensive energy consumption, the energy consumption allocation coefficient is used for conversion.

[0056] The calculation formula is: monthly operating hours of the equipment multiplied by unit energy consumption, then multiplied by the latest unit price of fuel and power, automatically yielding the consumption amount. When the same equipment consumes multiple energy sources such as water, electricity, natural gas, and nitrogen, each is calculated independently. This processing method resolves the issue of incompatibility between different types of energy consumption data formats.

[0057] For direct material input costs, the project consumption cost is calculated by multiplying the quantity of materials input in the material requisition record by the real-time unit price, and the corresponding cost is reduced by the amount of waste material recovery.

[0058] Amortization of intangible assets and other expenses are accrued and collected monthly according to pre-set allocation rules and the proportion of related projects. Specifically, amortization of intangible assets is accrued monthly using the straight-line method, and other expenses are allocated to the corresponding projects based on the proportion of labor hours for each project.

[0059] It should be noted that all automatically calculated results retain both the machine-calculated value and the manually corrected value. When a user manually corrects an automatically calculated result, the original machine-calculated value is not overwritten, and the correction operation is independently written to the modification log (including the operation source and IP address), thus ensuring that every data change has complete traceability.

[0060] After the above-mentioned allocation calculations are completed, the data flow enters the four-caliber parallel mapping link. The four-caliber parallel mapping link aggregates the expenses of each item by project and fills them into the corresponding standard auxiliary ledger tables. Then, the standard auxiliary ledger tables are summed to generate a summary of R&D expenditures. Finally, the summary of R&D expenditures calculates and generates a detailed list of R&D expense discounts according to the additional deduction rules. The entire process is summarized level by level along the path of expense items, standard auxiliary ledger tables, R&D expenditure summary, and R&D expense discount details. Specifically, the expense items include the following 11 categories: personnel labor costs, fuel and power costs, equipment depreciation costs, testing and inspection costs, maintenance and repair costs, external personnel labor costs, raw and auxiliary material costs, intangible asset amortization costs, new product design costs, outsourced R&D costs, and other costs. The standard auxiliary ledger tables at least include a raw and auxiliary material cost table, a fuel and power consumption summary table, an equipment depreciation cost summary table, an intangible asset amortization cost summary table, and other cost summary tables.

[0061] In the four-caliber parallel mapping link, based on four different mapping rules, the data summarized at each level is simultaneously converted into four output structures under different regulatory requirements. The data is mapped and output according to the accounting caliber, the high-tech enterprise certification caliber, the additional deduction caliber, and the R&D statistics caliber, respectively, generating a summary table of expenses for each project and a detailed ledger of expenses for each project.

[0062] The mapping rules between the various categories include at least the following: for the same raw and auxiliary material expenses, the additional deduction category records the full amount based on the actual consumption, while the accounting category records it based on the quantity used; for outsourced R&D expenses, the additional deduction category only allows 80% of the actual payment amount to be recorded, while other categories record the full amount based on the actual payment amount.

[0063] For other expense items, the additional deduction is solved using algebraic methods to address circular dependencies. Let Y be the sum of the five expenses (personnel costs, direct input costs, depreciation, amortization of intangible assets, and new product design fees) for all R&D projects. According to the additional deduction policy, there is a circular dependency between the other expense limit X and the sum of the five expenses Y: X itself is part of the total additional deduction for R&D expenses, and the total additional deduction for R&D expenses includes X; therefore, X cannot be calculated at once using simple multiplication. The system uses algebraic transformations to make the above circular dependency explicit, and solves it at once using the following formula:

[0064] X = Y × 10% ÷ (1 − 10%) = Y / 9

[0065] The system extracts the five expense data items for each project from the collected standard auxiliary ledger tables and calculates their sum Y. Then, it calculates the other expense limit X for all projects using the above formula. Next, it compares the actual total other expense amount for all projects with the limit X. If the actual total other expense amount for all projects does not exceed the limit X, the amount of other expenses that can be included in the additional deduction scope is the actual total other expense amount for all projects. If the actual total other expense amount for all projects exceeds the limit X, the amount of other expenses that can be included in the additional deduction scope is the limit X. If the actual total other expense amount for all projects exceeds the limit X, the system allocates the limit X to each project according to the proportion of each project's actual other expense amount to the total actual other expense amount for all projects. Finally, the above calculation results are written back to the additional deduction expense details for the corresponding project, and a consistency check is performed to ensure that the sum of the allocated amounts for each project equals the limit X. Through the above calculation process, the system explicitly transforms the implicit circular dependencies in the policy rules into executable algebraic solution steps, enabling the computer system to automatically complete the calculation and allocation of other expense limits.

[0066] Furthermore, the R&D expenditure summary includes built-in fields for calculating the additional deduction rate, additional deduction amount, and income tax reduction / exemption amount; the R&D expense preferential details table aligns with the tax declaration details structure, including at least the following declaration-level fields: wages and salaries, social insurance and housing fund contributions, direct consumption materials / fuel / power, intermediate test molds and process equipment, samples and prototypes that do not constitute fixed assets, trial product inspection fees, instrument and equipment operation and maintenance, operating lease equipment rental fees, instrument depreciation, equipment depreciation, amortization of software / patent rights / non-patented technologies, new product design fees, new process specification formulation fees, clinical trial fees / field trial fees, technical books and materials / translation / expert consultation / R&D insurance, etc. Preferably, the summary calculation of all R&D data for the current year is completed within seconds; the summary table of expenses for each project and the detailed ledger of expenses for each project support display by a single caliber, or simultaneously display by four calibers, and support setting a default display caliber.

[0067] Through the aforementioned four-caliber parallel mapping link, the same set of raw data is automatically converted into four different regulatory calibers. The differences between the calibers can be directly traced back to the mapping rule level, making the four sets of output data comparable and traceable.

[0068] S4. Abnormal risk identification and real-time intelligent early warning;

[0069] During data acquisition and allocation calculation, the system performs a two-layer anomaly check. The two-layer anomaly check and the four-caliber parallel mapping link described in S3 form a quality cascade constraint relationship. The first-layer check acts as the entry gate of the four-caliber mapping link, verifying the credibility of the data source entering the mapping link; the second-layer check acts as the exit filter of the four-caliber mapping link, verifying the structural rationality of the data output by each caliber of the mapping link.

[0070] The first layer is basic data logic verification, which starts during the data acquisition phase and identifies logical conflicts in the raw data in real time. This layer identifies two main types of anomalies: first, logical conflicts such as R&D hours exceeding the employee's total monthly attendance hours or equipment R&D hours exceeding the equipment's total monthly uptime, triggering an overtime warning; second, the completion progress of monthly R&D process documents and cost collection forms, pushing work progress warnings to projects lagging behind the planned nodes based on milestone nodes set in each project initiation report or project management module. The overtime warning specifically marks the anomaly type, the personnel or equipment involved, the corresponding project, and a comparison of the anomaly value with the baseline value; the work progress warning specifically displays the completion status of each project's monthly R&D materials, including submitted, under review, or incomplete. When raw data fails the first layer verification, the data is marked as unreliable and prevented from flowing into the four-caliber parallel mapping link described in S3. The system pushes a warning and awaits manual handling. Only when the raw data passes the first layer verification is it allowed to enter the four-caliber parallel mapping link for allocation calculation and multi-caliber output. This constitutes the first layer of verification as a quality constraint mechanism for the mapping link entry gate, avoiding the systemic risk that all four sets of data would output unreliable data due to source data distortion.

[0071] The second layer is cost structure verification, which is initiated after the allocation calculation and four-caliber mapping are completed to verify the rationality of the cost structure in the final settlement results. This layer compares the proportion of each item's cost category with a preset reasonable proportion threshold range. The preset reasonable proportion threshold range is set based on the industry's dynamic benchmark range for R&D expenses, and allows setting minimum and maximum proportions for each cost category by organization, project, and year / month. When the proportion of a certain category exceeds the threshold range, the system triggers a cost structure warning. Before cost structure verification, the base amount for verification is determined according to the "whether to exclude waste recycling amount" switch in the category proportion configuration rules, so that the same data can be verified independently with or without deducting waste recycling amount.

[0072] When the output data of a certain caliber triggers a cost structure warning during the second-level verification, the system performs tiered handling based on the severity of the warning: If only a single caliber shows a structural anomaly, the output of that caliber is marked as "requiring manual review," while the outputs of other calibers are output normally, achieving differentiated processing for each caliber and avoiding the blockage of all four outputs due to a single caliber anomaly; if multiple calibers show structural anomalies simultaneously, the system determines that it may stem from a deviation in the allocation calculation logic and triggers the S5 correction and recalculation process. All warning information is pushed to the management terminal after being labeled with the anomaly type, the involved projects, and the corresponding original documents.

[0073] Through the aforementioned layered verification architecture, the logical credibility is verified first during the data acquisition phase, and the structural rationality is judged after the calculation is completed, resulting in a significantly higher early warning accuracy rate than a single-layer verification scheme. Simultaneously, the second-layer verification results can be used to differentiate the outputs of the four sets of standards, enabling differentiated control over the quality of data outputs from different standards.

[0074] Once the warning is triggered, the data processing flow enters the S5 processing and optimization phase.

[0075] S5, abnormal closed-loop handling, dynamic optimization of mapping parameters and monthly data finalization;

[0076] After receiving the early warning information, the management system uses the source tracing function to view the original business documents associated with the abnormal data to locate the source of the problem and distinguish whether it is caused by incorrect selection of data collection criteria, manual entry errors, or deviations in allocation parameters. The system then issues rectification instructions to the corresponding responsible departments.

[0077] After the responsible department completes the data correction and re-uploads it, the recalculation process is triggered. The corrected data re-enters the four-caliber parallel mapping link described in S3 for recalculation, and the cost summary table is updated based on the corrected data. The description of the anomaly, the investigation process, the rectification plan, and the corrected data are fully archived to form an anomaly handling file.

[0078] In addition to the aforementioned single-correction recalculation mechanism, the system also has a built-in mapping parameter optimization engine that continuously monitors the early warning output of the second-level verification and automatically optimizes the mapping parameters of each expense item under each scope based on the early warning mode.

[0079] The engine abstracts the mapping rules into parameterized functions, taking the historical expense item percentage sequence, deviation direction, deviation magnitude, and warning trigger records as input datasets. With the goal of converging the mapping results toward the compliance threshold range, it automatically calculates the optimal inclusion ratio of the item under the corresponding caliber.

[0080] Specifically, the proportion is a free variable, ranging from 0 to 1. The optimization objective is to minimize the sum of squared weighted deviations between the mapping proportions of each historical period and the boundaries of the compliance threshold interval, i.e.:

[0081]

[0082] in, For the first The deviation between the mapping ratio of a historical period and the boundary of the compliance threshold range is calculated as follows: when the mapping ratio of the period is within the compliance threshold range, the deviation is recorded as zero; when the mapping ratio is lower than the lower limit, the deviation is the lower limit minus the mapping ratio; when the mapping ratio is higher than the upper limit, the deviation is the mapping ratio minus the upper limit. For the first The weighting coefficients for each historical period are determined using an exponential decay strategy:

[0083]

[0084] in Indicates the first The duration of the historical cycle from the current cycle, The attenuation rate constant ( >0). This weighting strategy gives higher weight to recent warning data in the optimization process, and the older the data, the smaller its impact on the current mapping parameters, thus ensuring the timeliness of the mapping parameters.

[0085] The aforementioned univariate convex optimization problem is solved using the golden section search or gradient descent method. When multiple candidate optimal parameters exist, the system selects the parameter that minimizes the rate of change of the inclusion ratio to maintain the continuity of data processing. After calculating the optimal inclusion ratio, the system uses the original data from each historical period as input and substitutes the optimized inclusion ratio to recalculate the mapping output. If the mapping output structure converges to the compliance threshold range before optimization, the optimization is confirmed as effective, and the optimized inclusion ratio is written into the mapping parameter table, taking effect in the mapping link of the next calculation cycle; if the mapping output structure does not converge or new structural deviations occur, the optimization result is abandoned and manual review is triggered. All mapping parameter optimization records are written to the mapping parameter change log, which includes at least the optimization time, optimization subject, coefficient before optimization, coefficient after optimization, comparison of the sum of squares of deviations before and after optimization, triggered warning mode, and effect verification status, ensuring the traceability of the mapping parameter evolution process.

[0086] The technical essence of this parameter optimization mechanism lies in: automatically generating the optimal inclusion ratio of the subject under this caliber by minimizing the objective function based on historical data, realizing the leap from rule matching to parameter generation, and enabling the mapping rule to have true data-driven adaptive capability.

[0087] After confirming that all data for the month is correct, monthly data locking is performed. Monthly data locking is executed by organizational unit and year / month; once locked, modifications to all measurement and summary data for that month are controlled. Combined with the modification logs of each of the aforementioned stages, the system forms a complete audit chain from initial data collection to final version approval, ensuring that the final output of R&D expense declaration data has reliable final version validity with traceable sources, trackable destinations, controllable nodes, and accountability.

[0088] Furthermore, the method includes a visualization step, displaying the completed cost data in the form of a cost structure ring diagram, monthly cost curves, and project status distribution maps. Content related to approval processes or process engines is only mentioned as the original data source in the aforementioned data collection steps, without elaborating on the configuration-driven or node state machine mechanisms within the process engine. The method also further includes a data closed loop for R&D input and output within the same platform: from the project initiation, through the association of projects, personnel, and equipment, as well as process measurement, monthly summaries and locking, standard reports and application outputs, to R&D results (patents / software copyrights / standards) and results transformation records, all data is progressively connected along the entire project lifecycle.

[0089] Example 2

[0090] Please see Figure 2 As shown, this embodiment provides a multi-caliber intelligent settlement and anomaly collection risk early warning system for R&D expenses, including the following modules:

[0091] The data acquisition and cleaning module is configured to automatically connect to the OA system, ERP system, R&D management system, and equipment monitoring system via API interfaces to retrieve personnel attendance and salary information, equipment ledgers and operation records, raw and auxiliary material requisition and waste recycling records, fuel and power consumption data, intangible asset amortization information, and outsourced contract execution plans, respectively. For offline paper documents not connected to the business systems, data is collected manually. This module also supports single entry and batch import of attendance records and requisition records, and can collect rest day configuration information. The collected raw data is sent to a data standardization and cleaning process. Based on the project number, R&D cycle, and R&D scope contained in the R&D project initiation report as the matching benchmark, operating expenses unrelated to R&D activities and expenditure records occurring outside the R&D cycle are removed. The field format, unit of measurement, and statistical caliber of each data source are standardized, and missing project affiliation identifiers are filled in based on the project binding information in the attendance records, establishing a link between each cost expenditure and the corresponding R&D project. After automatic cleaning and matching are completed, the results are pushed to the manual review end. R&D personnel and finance personnel jointly verify the automatic matching results. Once the review is passed, the results are locked into the standardized accounting basic database.

[0092] The settlement and mapping module is configured to: based on the standardized accounting database, call differentiated calculation models according to different expense categories to complete the collection of R&D expenses for each project. For personnel labor costs and equipment depreciation costs, they are allocated to the corresponding projects based on the proportion of man-hours invested by personnel or equipment in each R&D project; for power and energy consumption costs, it is compatible with both direct metering from instrument readings and conversion based on unit energy consumption coefficients, and is calculated independently when the same equipment consumes multiple energy sources; for direct input material costs, the corresponding cost is reduced by the amount of waste material recovery; for intangible asset amortization and other expenses, they are accrued and collected monthly according to preset allocation rules. This module retains both machine-calculated and manually corrected values ​​for all automatically calculated results, and independently writes manual correction operations into a modification log. After the above allocation calculation is completed, the data flow enters the four-caliber parallel mapping link, which aggregates the expenses of each item by project and fills them into the corresponding standard auxiliary ledger table. This process generates a summary of R&D expenditures and details of R&D expense preferential treatment at each level. Based on four different mapping rules, it is simultaneously converted into four output structures: accounting, high-tech enterprise certification, additional deduction, and R&D statistics, generating a summary table of expenses for each project and a detailed ledger of expenses for each project. For other expense items, the additional deduction caliber performs algebraic solving for circular dependencies. Through algebraic transformation, the implicit circular dependencies in the policy rules are made explicit into executable solution steps, automatically completing the calculation and allocation of other expense limits.

[0093] The verification and early warning module is configured to perform a two-layer anomaly verification during data acquisition and allocation calculation, forming a quality cascade constraint relationship with the four-caliber parallel mapping link. The first layer is basic data logic verification, which starts during the data acquisition phase. It identifies logical conflicts such as R&D hours exceeding the employee's total monthly attendance hours and equipment R&D hours exceeding the equipment's total monthly uptime, and triggers an overtime warning. Simultaneously, it tracks the completion progress of the monthly R&D process documents and cost collection forms and pushes work progress warnings. Data that fails the first layer verification is marked as untrusted data and prevented from flowing into the four-caliber parallel mapping link. The second layer is cost structure verification, which starts after the allocation calculation and four-caliber mapping are completed. It compares the cost item proportion of each project with a preset reasonable proportion threshold range. When the proportion of a certain item exceeds the threshold range, a cost structure warning is triggered. When the output data of a certain standard triggers a cost structure warning, if only a single standard shows a structural anomaly, the output of that standard is marked as "requires manual review," while other standards output normally. If multiple standards show structural anomalies simultaneously, an anomaly closed-loop handling is triggered. All warning information is pushed to the management terminal after being labeled with the anomaly type, the involved projects, and the corresponding original documents.

[0094] The closed-loop and optimization module is configured as follows: After receiving the early warning information, the management terminal uses the penetrating source tracing function to view the original business documents associated with the abnormal data to locate the source of the problem, issues rectification instructions to the corresponding responsible departments, and after the responsible departments complete the data correction and re-upload it, the recalculation process is triggered. The corrected data re-enters the four-caliber parallel mapping link to perform recalculation, and the cost summary table is updated according to the corrected data. The description of the abnormal problem, the investigation process, the rectification plan and the corrected data are completely archived to form an abnormal handling file. This module also has a built-in mapping parameter optimization engine that continuously monitors the early warning output of the second-level verification. Based on the early warning mode, it automatically optimizes the mapping parameters of each expense item under each caliber: the mapping rules are abstracted into parameterized functions, with the goal of converging the mapping results towards the compliance threshold range. By minimizing the sum of squared weighted deviations between the mapping proportions of each historical period and the boundary of the compliance threshold range, the optimal inclusion ratio of the item under the corresponding caliber is automatically calculated. The weights of each period adopt an exponential decay strategy, with higher weights for recent data and lower weights for older data. This univariate convex optimization problem is solved using the golden section search or gradient descent method. After calculating the optimal inclusion ratio, it is verified using the original data of each historical period as input. If the mapping output structure converges towards the compliance threshold range compared to before optimization, the optimized inclusion ratio is written into the mapping parameter table and takes effect in the mapping link of the next calculation period. All mapping parameter optimization records are written into the mapping parameter change log.

[0095] The finalization and auditing module is configured to: lock the monthly data after confirming that all data for the month is correct, by organizational unit and year and month. After locking, all modifications to the measurement and summary data for the month are controlled. Combined with the modification logs of the aforementioned stages, a full audit chain is formed from the original data collection to the final version, so that the final output R&D expense declaration data has a credible finalization effect with traceable source, traceable destination, controllable nodes, and accountability.

[0096] The display and closed-loop management module is configured to: display the completed cost data in the form of a cost structure ring diagram, monthly cost curves, and project status distribution map; realize the data closed loop of R&D input and R&D output within the same platform; from the start of the R&D project, through the association of projects, personnel and equipment, as well as process measurement, monthly summary and locking, standard reports and application output, to the R&D results and results transformation records, all data are connected step by step along the entire life cycle of the project.

[0097] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for intelligent calculation and anomaly collection risk early warning of R&D expenses using multiple metrics, characterized in that: include: Collect raw R&D data from various business systems of the enterprise, perform standardized cleaning and project matching audit on the raw R&D data, establish a link between each cost expenditure and the corresponding R&D project, and lock it into a standardized accounting basic database; Based on the standardized accounting database, differentiated allocation calculation models are called according to different expense categories to collect the R&D expenses of each project, and the collected expense data of each category is obtained. The collected expense data of each subject is filled into the corresponding standard auxiliary ledger table, and the R&D expenditure summary and R&D expense discount details are generated level by level. Based on four different mapping rules, it is simultaneously converted into four output structures: accounting caliber, high-tech enterprise certification caliber, additional deduction caliber and R&D statistical caliber, generating a summary table of expenses for each project and a detailed ledger of expenses for each project. During the data acquisition and allocation calculation process, a two-layer anomaly check is performed. The first layer of check acts as the entry gate for the four sets of mapping rules, verifying the credibility of the data source entering the mapping link. Data that fails the first layer of check is prevented from flowing into the mapping link. The second layer of check acts as the exit filter for the mapping link, verifying the structural rationality of the data output by the mapping link. When an anomaly is triggered, a correction and recalculation process is executed. The corrected data is re-entered into the mapping link for recalculation. Based on historical early warning data, the mapping parameters of each expense item under each caliber are optimized. After confirming that all data for the month is correct, monthly data locking is executed, forming a complete audit chain.

2. The method for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses according to claim 1, characterized in that, Standard auxiliary ledgers should include at least the following: raw and auxiliary material expense statement, fuel and power consumption summary statement, equipment depreciation expense summary statement, intangible asset amortization expense summary statement, and other expense summary statement. The four different mapping rules include at least the following: for the same raw and auxiliary material expenses, the additional deduction is recorded in full based on the actual consumption, while the accounting method records it based on the quantity used. For outsourced R&D expenses, the additional deduction is only allowed at 80% of the actual payment amount, while other methods allow for the full amount of the actual payment amount to be deducted.

3. The method for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses according to claim 1, characterized in that, For other expense items, the additional deduction method is solved using algebraic methods to address circular dependencies: The five expenses of all R&D projects are summed up. These five expenses include personnel costs, direct input costs, depreciation costs, amortization of intangible assets, and new product design fees. According to the additional deduction policy, there is a circular dependency relationship between the other expense limit and the sum of the five expenses. The circular dependency relationship is transformed into a form that can be solved in one go through algebraic transformation to calculate the other expense limit for all projects. The actual total amount of other expenses incurred by all projects is compared with the limit. When the actual total amount incurred exceeds the limit, the limit is allocated to each project according to the proportion of the actual amount of other expenses incurred by each project to the total amount of other expenses incurred by all projects, and a consistency check is performed.

4. The method for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses according to claim 1, characterized in that, The first layer of verification is a logical consistency verification, which is initiated during the data acquisition process. It identifies logical conflict situations such as R&D hours exceeding the employee's total monthly attendance hours and equipment R&D hours exceeding the equipment's total monthly power-on time, and triggers an overtime warning. The second layer of verification is the cost structure verification, which is initiated after the allocation calculation and four sets of mapping are completed. It compares the proportion of each item's cost items with the preset reasonable proportion threshold range. When the proportion of a certain item exceeds the threshold range, a cost structure warning is triggered.

5. The method for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses according to claim 4, characterized in that, When the output data of a certain caliber triggers a cost structure warning, if only a single caliber shows a structural anomaly, the output of that caliber will be marked as requiring manual review, while other calibers will output normally. If structural anomalies occur simultaneously in multiple calibers, a correction and recalculation process will be triggered.

6. The method for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses according to claim 1, characterized in that, Optimize the mapping parameters of each expense item under each scope, specifically as follows: The mapping rules are abstracted into parameterized functions, with the optimization objective of minimizing the weighted sum of squared deviations between the mapping ratio of each historical period and the boundary of the compliance threshold range, and the optimal inclusion ratio of the item under the corresponding caliber is automatically calculated. The deviation value for each historical period is determined as follows: When the mapping ratio of a period is within the compliant threshold range, the deviation is zero; when it is below the lower limit, the deviation is the lower limit minus the mapping ratio; when it is above the upper limit, the deviation is the mapping ratio minus the upper limit. The weight of each period is determined using an exponential decay strategy, with recent data having a higher weight than older data.

7. The method for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses according to claim 1, characterized in that, Monthly data locking is performed by organizational unit and year / month. Once locked, modifications to all measurement and summary data for that month are controlled. The entire audit chain consists of modification logs for each of the aforementioned stages.

8. The method for intelligent multi-caliber calculation and anomaly collection risk early warning of R&D expenses according to claim 1, characterized in that, It also includes visualizations, including: The completed cost data will be displayed in the form of a cost structure ring diagram, monthly cost curves, and project status distribution map; it also includes a data loop of R&D input and output: from the start of the R&D project, through the association of project, personnel and equipment, as well as process measurement, monthly summary and locking, standard reports and application output, to the R&D results and results transformation records, all data are connected step by step along the entire life cycle of the project.

9. A multi-caliber intelligent settlement and anomaly collection risk early warning system for R&D expenses, characterized in that: include: The data collection and cleaning module is configured to collect raw R&D data from various business systems of the enterprise, perform standardized cleaning and project matching audit on the raw R&D data, establish a link between each cost expenditure and the corresponding R&D project, and lock it into a standardized accounting basic database. The settlement and mapping module is configured to, based on the standardized accounting database, call different allocation calculation models according to different expense items to complete the collection of R&D expenses for each project, fill the collected expense data of each item into the corresponding standard auxiliary ledger, and generate a summary of R&D expenditures and details of R&D expense discounts by level. According to four different mapping rules, it is simultaneously converted into four output structures: accounting caliber, high-tech enterprise certification caliber, additional deduction caliber, and R&D statistical caliber, generating a summary table of expenses for each project and a detailed ledger of expenses for each project. The verification and early warning module is configured to run a two-layer anomaly verification during the data acquisition and allocation calculation process. The first layer of verification acts as the entry gate for the four sets of mapping rules, verifying the credibility of the data source entering the mapping link. Data that fails the first layer of verification is prevented from flowing into the mapping link. The second layer of verification acts as the exit filter for the mapping link, verifying the structural rationality of the data output by the mapping link. The closed-loop and optimization module is configured to execute a correction and recalculation process when an anomaly is triggered. The corrected data is re-entered into the mapping link for recalculation, and the mapping parameters of each expense item under each caliber are optimized based on historical early warning data. The version finalization and auditing module is configured to lock the monthly data after confirming that all data for the month is correct, thus forming a complete audit chain.

10. The R&D expense multi-caliber intelligent settlement and anomaly collection risk early warning system according to claim 9, characterized in that, In the closed-loop and optimization module, the mapping parameters of each expense item under each caliber are optimized. Specifically, the mapping rules are abstracted into parameterized functions. The optimization objective is to minimize the weighted sum of squared deviations between the mapping ratio of each historical period and the boundary of the compliance threshold interval. The optimal inclusion ratio of the item under the corresponding caliber is automatically calculated. The deviation value for each historical period is determined as follows: when the mapping ratio of the period is within the compliance threshold range, the deviation is zero; when it is below the lower limit, the deviation is the lower limit minus the mapping ratio; when it is above the upper limit, the deviation is the mapping ratio minus the upper limit. The weight of each period is determined using an exponential decay strategy, with recent data having a higher weight than older data.

Images