Science and technology project performance dynamic prediction method and system combining digital twinborn and artificial intelligence

Abstract

The invention provides a science and technology project performance dynamic prediction method and system combining digital twinning and artificial intelligence, and relates to the technical field of science and technology project management data analysis. According to the method, a performance indicator system and project process data are obtained, a mapping relation between indicator calibers and process data is established, a project digital twinborn body containing task dependence, milestone and resource boundary constraints is constructed, and a feature sequence with consistent constraints is generated; on the basis, a consistency constraint training performance prediction model is introduced, rolling prediction and confidence evaluation in the operation period are achieved, calibration of the model or the mapping relation is triggered when prediction credibility is insufficient or management constraint is violated, and therefore the caliber consistency, constraint rationality and trend availability of a performance prediction result are improved.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology for science and technology project management, and in particular to a method and system for dynamic prediction of science and technology project performance that combines digital twins and artificial intelligence. Background Technology

[0002] In research institutes, universities, and corporate R&D systems, performance management of science and technology projects commonly employs a combination of full-process management and milestone acceptance. Management practices typically rely on project mandates, contractual targets, annual plans, and budgets, resulting in supporting records such as progress updates, phase reports, budget execution, results registration, intellectual property management, and risk and issue ledgers. Performance conclusions are reached at stages such as project approval, mid-term review, and final acceptance. Regarding evaluation methods, a common approach is a comprehensive scoring system based on indicators, combined with expert review opinions. Some institutions introduce earned value management (EVM) and other management methods to reflect deviations in progress and funding. While these methods support performance assessment and compliance management, performance conclusions often depend on interim materials and milestone evaluations, making it difficult to continuously depict the changing trends of project performance over time.

[0003] In recent years, the digitalization of scientific research management has led to more continuous and granular project process data, covering aspects such as project plan adjustments, collaborative records, funding voucher transfers, and results formation and transformation. This makes dynamic analysis based on process data a feasible direction. Meanwhile, advances in artificial intelligence research in time series modeling, anomaly detection, and uncertainty assessment have gradually drawn attention to using historical project process data for performance trend prediction. On the other hand, digital twins have expanded from traditional physical objects such as equipment and production lines to management objects of R&D activities. By unifying state representation and constraint relationships, they depict the correlation between project progress, resource input, and output, supporting simulation and trend analysis. Combining the state representation of digital twins with the predictive capabilities of artificial intelligence is becoming a research direction for dynamic prediction and process control of scientific and technological project performance.

[0004] Existing solutions still have significant shortcomings in implementation. First, the definitions of indicators, statistical methods, and data collection rules vary considerably among different units or project categories. Inconsistencies often arise between the data used for performance evaluation and the data accumulated by the process system, making it difficult for prediction models trained on historical data to stably adapt to actual evaluation rules. Second, when using data-driven models for prediction, it is often difficult to explicitly incorporate management constraints such as task dependencies, milestone constraints, and funding and resource boundaries. The prediction results may not match the actual execution patterns of the project, affecting interpretability and credibility. Third, when project management rules are adjusted, the external environment changes, or data distribution drifts, there is a lack of effective online consistency verification and calibration mechanisms. This causes the prediction accuracy and stability to decline over time, making it difficult to continuously support early warning and intervention in management decisions. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for dynamic prediction of the performance of science and technology projects that combines digital twins and artificial intelligence. Through performance caliber mapping, digital twin constraints, and consistency training mechanisms, the performance prediction results maintain caliber consistency and trend usability while meeting project management constraints.

[0006] To achieve the above objectives, the present invention provides the following solution: A method for dynamically predicting the performance of technology projects by combining digital twins and artificial intelligence, comprising: Obtain the performance indicator system and project process data of the target technology project, and determine the definition, statistical granularity, collection rules and evaluation cycle of each indicator in the performance indicator system, and establish the mapping relationship between the performance indicator system and the project process data fields; Based on the performance indicator system and project baseline data, a set of project twin state variables is defined, and task dependency constraints, milestone constraints, and resource boundary constraints are set for the set of project twin state variables to construct a project digital twin that can be updated with the project process data; The project process data is converted into a state sequence of the project twin state variable set according to the mapping relationship, and a constraint-consistent feature sequence is generated under the task dependency constraint, the milestone constraint, and the resource boundary constraint; The performance prediction model is trained using the feature sequence as input and the performance labels obtained by converting historical performance review results according to the definition, statistical granularity, aggregation rules and evaluation cycle as output. Consistency constraints are introduced to suppress inconsistencies between the prediction output and the task dependency constraints, milestone constraints and resource boundary constraints. During the project operation period, the project digital twin is updated based on the newly added project process data, and the feature sequence is updated. The updated feature sequence is input into the performance prediction model to output the performance prediction results and confidence level within the preset prediction window. When the confidence level is lower than the threshold or the constraint consistency check performed on the performance prediction result based on the task dependency constraint, the milestone constraint and the resource boundary constraint fails, the mapping relationship is updated or the parameters of the performance prediction model are adjusted, and subsequent dynamic performance prediction is performed using the updated mapping relationship or the adjusted performance prediction model.

[0007] Preferably, the process involves acquiring the performance indicator system and project process data of the target technology project, determining the definition, statistical granularity, aggregation rules, and evaluation cycle of each indicator in the performance indicator system, and establishing a mapping relationship between the performance indicator system and the project process data fields, including: Obtain the indicator description information that records the definition, statistical granularity, aggregation rules and evaluation cycle of each indicator in the performance indicator system; Based on the indicator description information, the definition, statistical granularity, aggregation rules, and evaluation period of each indicator in the performance indicator system are determined respectively; Based on the aggregation rules, the project process data fields corresponding to each indicator in the performance indicator system are determined in the project process data, and the time aggregation method of the project process data fields is determined based on the statistical granularity and the evaluation period to form the mapping relationship.

[0008] Preferably, the process further includes acquiring the performance indicator system and project process data of the target technology project, determining the definition, statistical granularity, aggregation rules, and evaluation cycle of each indicator in the performance indicator system, and establishing a mapping relationship between the performance indicator system and the project process data fields. The definitions of each indicator in the performance indicator system are made consistent in terms of dimensions and units of measurement. The statistical granularity of each indicator in the performance indicator system is determined to be one of daily, weekly, or monthly. The evaluation period for each indicator in the performance indicator system is determined as a continuous time interval composed of multiple statistical granularities, and the evaluation period is aligned with the planned time boundary in the project baseline data.

[0009] Preferably, a set of project twin state variables is defined based on the performance indicator system and project baseline data, including: Based on each indicator in the performance indicator system, define twin state variables that correspond one-to-one with each indicator; Based on the task decomposition structure, milestone plan, budget items and resource allocation in the project baseline data, define constraint state variables to characterize the task dependency constraints, the milestone constraints and the resource boundary constraints; The twin state variables and the constraint state variables are combined to form the project twin state variable set.

[0010] Preferably, task dependency constraints, milestone constraints, and resource boundary constraints are set for the project's twin state variable set, including: The task sequence relationship is determined based on the task decomposition structure in the project baseline data, and the task sequence relationship is used as the task dependency constraint. Based on the milestone plan in the project baseline data, milestone targets and milestone dates are determined, and the milestone targets and milestone dates are used as the milestone constraints. The funding cap or resource cap is determined based on the budget items and resource allocation in the project baseline data, and the funding cap or resource cap is used as the resource boundary constraint.

[0011] Preferably, converting the project process data into a state sequence of the project twin state variable set according to the mapping relationship includes: The project process data is divided into multiple time intervals according to the evaluation cycle. For the project process data within each time interval, determine the periodic values ​​of each twin state variable in the project twin state variable set according to the mapping relationship; The period values ​​of each time interval are arranged in chronological order to obtain the state sequence.

[0012] Preferably, generating a constraint-consistent feature sequence under the task dependency constraint, the milestone constraint, and the resource boundary constraint includes: Based on the task dependency constraints, the milestone constraints, and the resource boundary constraints, the consistency of the state sequence is determined to obtain the consistency determination result. The state values ​​in the state sequence that do not satisfy the task dependency constraint, the milestone constraint, or the resource boundary constraint are corrected so that the corrected state values ​​satisfy the task dependency constraint, the milestone constraint, and the resource boundary constraint. The corrected state sequence is determined as a characteristic sequence consistent with the constraints.

[0013] Preferably, a performance prediction model is trained using the feature sequence as input and performance labels obtained by converting historical performance review results according to the definition, statistical granularity, aggregation rules, and evaluation cycle as output. Consistency constraints are introduced to suppress inconsistencies between the prediction output and the task dependency constraints, milestone constraints, and resource boundary constraints, including: Based on the feature sequences in the training samples, the performance prediction model outputs the performance prediction results within the prediction window; Substituting the performance prediction results into the task dependency constraint, the milestone constraint, and the resource boundary constraint respectively, we obtain the task dependency inconsistency, milestone inconsistency, and resource boundary inconsistency. The consistency constraint penalty is determined based on the task dependency inconsistency, the milestone inconsistency, and the resource boundary inconsistency. The prediction error is determined based on the performance prediction results and the performance labels. The training objective is determined by the prediction error and the consistency constraint penalty, and the parameters of the performance prediction model are trained according to the training objective to reduce the consistency constraint penalty and the prediction error, thereby suppressing the inconsistency between the performance prediction result and the task dependency constraint, the milestone constraint and the resource boundary constraint.

[0014] Preferably, when the confidence level is lower than a threshold or the constraint consistency check performed on the performance prediction result based on the task dependency constraint, the milestone constraint, and the resource boundary constraint fails, the mapping relationship is updated or the parameters of the performance prediction model are adjusted, including: The performance prediction results are subjected to task dependency consistency test, milestone consistency test, and resource boundary consistency test, respectively. The task dependency consistency test is used to determine whether the performance prediction results meet the task sequence relationship defined by the task dependency constraint; the milestone consistency test is used to determine whether the performance prediction results meet the milestone achievement conditions defined by the milestone constraint; and the resource boundary consistency test is used to determine whether the performance prediction results meet the resource upper limit conditions defined by the resource boundary constraint. If any of the task dependency consistency test, milestone consistency test, and resource boundary consistency test fails to meet the corresponding constraint, the constraint consistency test is determined to be unsuccessful, and the mapping relationship is updated or the parameters of the performance prediction model are adjusted. When the task dependency consistency check, the milestone consistency check, and the resource boundary consistency check all satisfy the corresponding constraints, the constraint consistency check is determined to be passed.

[0015] A dynamic performance prediction system for technology projects that combines digital twins and artificial intelligence includes: The data mapping and indicator definition unit is used to acquire the performance indicator system and project process data of the target science and technology project, and to determine the definition, statistical granularity, collection rules and evaluation cycle of each indicator in the performance indicator system, and to establish the mapping relationship between the performance indicator system and the project process data fields. The twin construction and constraint setting unit is used to define a set of project twin state variables based on the performance indicator system and project baseline data, and to set task dependency constraints, milestone constraints and resource boundary constraints for the set of project twin state variables, so as to construct a project digital twin that can be updated with the project process data; The state sequence generation and feature extraction unit is used to convert the project process data into a state sequence of the project twin state variable set according to the mapping relationship, and generate a constrained feature sequence under the task dependency constraint, the milestone constraint and the resource boundary constraint. The consistency constraint model training unit is used to train a performance prediction model with the feature sequence as input and the performance labels obtained by converting historical performance review results according to the definition, the statistical granularity, the aggregation rule and the evaluation period as output. It introduces consistency constraints to suppress inconsistencies between the prediction output and the task dependency constraint, the milestone constraint and the resource boundary constraint. The dynamic update and performance prediction unit is used to update the project digital twin and the feature sequence based on the newly added project process data during the project operation period, and input the updated feature sequence into the performance prediction model to output the performance prediction results and confidence level within the preset prediction window; The model adaptive correction and optimization unit is used to trigger the updating of the mapping relationship or the adjustment of the parameters of the performance prediction model when the confidence level is lower than the threshold or the constraint consistency test performed on the performance prediction result based on the task dependency constraint, the milestone constraint and the resource boundary constraint fails, and then use the updated mapping relationship or the adjusted performance prediction model to perform subsequent dynamic performance prediction.

[0016] The present invention discloses the following technical effects: (1) This invention obtains the performance indicator system and project process data and determines the definition, statistical granularity, collection rules and evaluation cycle of each indicator. Then, it establishes a mapping relationship between the performance indicator system and the project process data fields, so that the performance evaluation caliber and the source of process data form a clear correspondence, reduce the performance deviation caused by the difference in statistical caliber of different links, and improve the consistency and comparability of the performance prediction input data.

[0017] (2) Based on the performance indicator system and project baseline data, this invention defines a set of project twin state variables and sets task dependency constraints, milestone constraints and resource boundary constraints. In this way, the digital twin carries the key states and management constraints in the project execution process, so that performance prediction no longer depends on the fitting of isolated historical data, but is constrained by the actual execution boundary conditions of the project, thereby improving the degree of fit of the prediction conclusion with the project management rules.

[0018] (3) The present invention converts project process data into a state sequence of twin state variable set according to the mapping relationship, and generates a feature sequence with consistent constraints under task dependency constraints, milestone constraints and resource boundary constraints, so that the feature expression entering the model is consistent with the project constraints, reduces the interference of process data fluctuations or abnormal records on prediction, and improves the ability of features to represent the evolution of the project's real state.

[0019] (4) This invention introduces consistency constraints when training the performance prediction model to suppress inconsistencies between the prediction output and task dependency constraints, milestone constraints and resource boundary constraints, so that the model output can approximate the historical performance review results under the premise of meeting management constraints, reduce the probability of prediction results that do not meet the task sequence relationship, milestone requirements or resource limit conditions, and improve the credibility and interpretability of the prediction results.

[0020] (5) During the project operation period, the present invention updates the digital twin of the project based on the newly added project process data and updates the feature sequence, outputs the performance prediction results and confidence level within the prediction window, and triggers the update of the mapping relationship or adjustment of the model parameters when the confidence level is lower than the threshold or the constraint consistency test fails, thereby forming a dynamic adaptive mechanism for data distribution changes and management caliber changes, and improving the prediction stability and continuous availability during long-term operation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart of the method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The purpose of this invention is to provide a method and system for dynamic prediction of the performance of science and technology projects that combines digital twins and artificial intelligence. Through constraint-consistent feature generation, model consistency suppression, and online calibration triggering mechanisms, the performance prediction output is made more in line with the project execution rules and the long-term prediction stability is improved.

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a method for dynamically predicting the performance of science and technology projects by combining digital twins and artificial intelligence, including: Step 100: Obtain the performance indicator system and project process data of the target technology project, and determine the definition, statistical granularity, collection rules and evaluation cycle of each indicator in the performance indicator system, and establish the mapping relationship between the performance indicator system and the project process data fields; Step 200: Define the project twin state variable set based on the performance indicator system and project baseline data, and set task dependency constraints, milestone constraints and resource boundary constraints for the project twin state variable set to build a project digital twin that can be updated with project process data; Step 300: Convert the project process data into a state sequence of the project twin state variable set according to the mapping relationship, and generate a constrained feature sequence under task dependency constraints, milestone constraints and resource boundary constraints; Step 400: Train the performance prediction model with the feature sequence as input and the performance labels obtained by transforming historical performance review results according to the definition, statistical granularity, aggregation rules and evaluation cycle as output. Introduce consistency constraints to suppress inconsistencies between the prediction output and task dependency constraints, milestone constraints and resource boundary constraints. Step 500: During the project operation period, update the project digital twin based on the newly added project process data and update the feature sequence. Input the updated feature sequence into the performance prediction model and output the performance prediction results and confidence level within the preset prediction window. Step 600: When the confidence level is lower than the threshold or the constraint consistency test performed on the performance prediction results based on task dependency constraints, milestone constraints and resource boundary constraints fails, the mapping relationship is updated or the parameters of the performance prediction model are adjusted, and the updated mapping relationship or the adjusted performance prediction model is used to perform subsequent dynamic performance prediction.

[0027] In this embodiment, step 100 first obtains the performance indicator system corresponding to the target science and technology project and the project process data generated during project execution. The performance indicator system can be derived from the performance indicator provisions in the project task book, performance evaluation methods, project approval documents, mid-term inspection requirements, or project completion acceptance standards. The project process data can include progress records, funding execution records, achievement registration records, and risk event records generated during project execution. To ensure consistency in performance evaluation criteria, this embodiment further obtains indicator description information to describe the performance indicator system. This indicator description information clarifies the definition, statistical granularity, aggregation rules, and evaluation cycle of each indicator in the performance indicator system. The indicator definition defines the performance meaning and value boundaries of each indicator. For example, the progress completion rate is defined as the ratio of completed tasks to planned tasks, and the expenditure execution rate is defined as the ratio of compliant expenditures to budgeted expenditures. The statistical granularity defines the time interval for updating indicator values, such as 7 days or 30 days. The collection rules define the scope of data that can be included in the indicator statistics, such as only including expenditure records with an audited status or only including records of accepted deliverables. The evaluation period defines the time interval boundaries corresponding to performance review and performance forecast, such as evaluating within a continuous 90-day interval or within a time interval divided by project phases. Through these definitions, each indicator has a unified technical meaning and statistical boundaries under different data source conditions.

[0028] After obtaining the indicator description information, this embodiment determines the definition, statistical granularity, aggregation rules, and evaluation period of each indicator in the performance indicator system based on the indicator description information, and uses the definition, statistical granularity, aggregation rules, and evaluation period as constraints for subsequently establishing mapping relationships. Specifically, for each indicator, this embodiment transforms the statistical method specified in the indicator description information into a unified time aggregation rule, enabling project process data to be aggregated under the same statistical boundaries. For example, when the statistical granularity of an indicator is set to 30 days, this embodiment aggregates project process data according to consecutive 30-day intervals; when the evaluation period is set to 90 days, this embodiment includes three consecutive statistical granularity intervals in the same evaluation period, thereby ensuring the comparability of indicator values ​​within the evaluation period. For progress indicators, a shorter statistical granularity can be used to reflect changes in execution pace; for funding indicators, a longer statistical granularity can be used to reflect trends in fund usage; for phased achievement indicators, the evaluation period can be consistent with the project milestone stage. By using the above methods, different types of indicators are matched with their management attributes in terms of time scale, avoiding the distortion of performance judgment due to inconsistent statistical periods.

[0029] Based on this, this embodiment determines the project process data fields corresponding to each indicator in the performance indicator system in the project process data according to the collection rules, and determines the time merging method of the project process data fields according to the statistical granularity and the evaluation period, thereby forming a mapping relationship between the performance indicator system and the project process data fields. The mapping relationship is used to limit the correspondence between indicators and project process data, and includes at least the data field range corresponding to the indicator and the time merging boundary. For example, for the budget execution rate indicator, this embodiment determines the budget amount field and the compliant expenditure amount field in the project process data as corresponding fields, and limits the compliant expenditure amount to records whose occurrence date falls within the evaluation period and has been approved before merging; for the progress completion rate indicator, this embodiment determines the total number of planned tasks field and the number of completed tasks field as corresponding fields, and limits the completed tasks to records whose completion date falls within the statistical granularity range and whose status is completed before merging; for output-type indicators, this embodiment determines the result category field and the result quantity field in the result registration record as corresponding fields, and limits the result category to the result type consistent with the project task book before statistics are performed. Through the above mapping relationship, project process data can be uniformly categorized into the effective value basis of each indicator under the constraints of the statistical granularity and the evaluation period, thereby providing clear, consistent and reproducible data support for the subsequent construction of the project digital twin and the formation of the state sequence.

[0030] In this embodiment, as a preferred implementation of step 100, after obtaining the performance indicator system and project process data of the target technology project, this embodiment further performs consistency processing on the various indicators in the performance indicator system to ensure that data from different sources have a unified technical meaning in statistical analysis and evaluation. Specifically, this embodiment first unifies the dimensions and units of measurement involved in the definition of each indicator. Dimensions are used to characterize the physical or managerial meaning category of the indicator, and units of measurement are used to limit the numerical expression of the indicator. For example, funding-related indicators are uniformly limited to RMB as the unit of measurement, time-related indicators are uniformly limited to days as the unit of measurement, and percentage-type indicators are uniformly limited to percentages, thereby avoiding incomparability of statistical results due to differences in units. Based on this, this embodiment determines the statistical granularity of each indicator in the performance indicator system to be one of daily, weekly, or monthly, so that the indicator values ​​have a clear update frequency on the time scale. For example, the statistical granularity of progress-type indicators is set to 7 days, and the statistical granularity of funding-type indicators is set to 30 days, to match the sensitivity of different indicators to process changes. Furthermore, this embodiment defines the evaluation period for each indicator in the performance indicator system as a continuous time interval composed of multiple statistical granularities, and aligns the evaluation period with the planned time boundary in the project baseline data. For example, three consecutive statistical granularity intervals constitute one evaluation period, or the evaluation period is set to 180 days and consistent with the start and end times of the project phase plan. Through this approach, the definition, statistical granularity, and evaluation period of each indicator are consistent in terms of dimensions, time scale, and planned boundaries, thereby providing a clear, reproducible, and numerically bounded technical foundation for establishing the mapping relationship between the subsequent performance indicator system and project process data fields.

[0031] In this embodiment, step 200 is used to construct the state representation basis of the project's digital twin under the constraints of the performance indicator system and project baseline data. This embodiment first defines twin state variables corresponding one-to-one with each indicator in the performance indicator system. These twin state variables characterize the observable performance state of the project during execution, maintaining consistent meaning and time scale with the indicators. For example, when the performance indicator system includes indicators such as progress completion rate, budget execution rate, output quantity, and risk handling timeliness, this embodiment defines progress state variables, budget state variables, output state variables, and risk state variables respectively, ensuring that each twin state variable maintains the same expression as its corresponding indicator. This guarantees that subsequent state evolution and performance prediction can directly express changes in project performance using the twin state variables as the carrier.

[0032] Building upon the defined twin state variables, this embodiment further defines constraint state variables based on the task decomposition structure, milestone plan, budget items, and resource allocation in the project baseline data. These constraint state variables are used to explicitly represent the constraints during project execution, enabling them to form a unified state space together with the twin state variables. For example, this embodiment extracts task nodes and their sequential relationships from the task decomposition structure to form constraint state variables representing task dependencies; extracts stage goals and date requirements from the milestone plan to form constraint state variables representing milestone constraints; and extracts stage budget caps and key resource caps from the budget items and resource allocation to form constraint state variables representing resource boundaries. By transforming constraints into state variables that can be updated over time and referenced in consistency checks, the reasonableness of subsequent prediction outputs can be evaluated within the constraint framework.

[0033] Building upon the above, this embodiment combines twin state variables and constraint state variables to form a project twin state variable set, and constructs a project digital twin that can be updated with project process data. The project digital twin is used to represent the project's state values ​​and their constraints across different evaluation periods, allowing project process data to continuously drive updates to the twin state variables, while constraint state variables limit the feasible range of the updated state. To ensure sufficient disclosure, this embodiment provides an operational, quantifiable boundary example: the task decomposition structure in the project baseline data contains no fewer than 20 task nodes; the milestone plan contains no fewer than 3 milestone nodes; the budget items include at least two of the following: personnel, equipment, materials, and testing; resource allocation includes at least one type of key equipment resource and one type of key personnel resource, thus ensuring that the project digital twin has a clear state and constraint scale, preventing it from being unimplementable due to abstract descriptions.

[0034] Furthermore, this embodiment sets task dependency constraints, milestone constraints, and resource boundary constraints for the project's digital twin state variable set to form a constraint system that can be used for consistency judgment and constraint consistency verification. Specifically, this embodiment determines the task sequence based on the task decomposition structure and uses the task sequence as a task dependency constraint. For example, it limits the start time of task 2 to no earlier than the completion time of task 1, or limits the completion of at least 60% of the preceding tasks before a subsequent task can be considered complete. This embodiment determines milestone targets and milestone dates based on the milestone plan and uses the milestone targets and milestone dates as milestone constraints. For example, it limits the first milestone date to 90 days after project start and the corresponding output quantity to 3 items. This embodiment determines the funding limit or resource limit based on budget items and resource allocation and uses the funding limit or resource limit as a resource boundary constraint. For example, it limits the funding execution rate to no more than 100% and the key equipment occupancy time to no more than 160 hours within the evaluation period. Through the above constraint settings, the project digital twin has clear feasible boundaries in the subsequent update and prediction process, providing sufficient support for consistency constraint training and runtime consistency verification.

[0035] In this embodiment, to ensure that the project digital twin can be updated along with the project process data and that the update results satisfy task dependency constraints, milestone constraints, and resource boundary constraints, this embodiment describes the update of the project twin state variable set in each evaluation cycle as a form of "predictive update plus constraint projection": in, For the first The state vector of the project twin state variable set for each evaluation period; This is the state vector from the previous evaluation period; For the first Each evaluation cycle is an observation vector obtained from project process data according to a mapping relationship; This is a mapping matrix used to map the observation vector to the state vector; The constraint projection operator is used to project the update result within the parentheses onto the feasible region. The feasible region is defined by task dependency constraints, milestone constraints, and resource boundary constraints.

[0036] In this embodiment, to ensure that the three types of constraints can be directly referenced in consistency judgment and subsequent consistency verification, the task dependency constraints, milestone constraints, and resource boundary constraints are uniformly represented as a set of feasible constraints on the state vector, and a constraint violation degree is introduced to characterize the degree of constraint compliance of the state update or prediction output: in, State vector The degree of constraint violation; Let be a vector of constraint functions consisting of task dependency constraints, milestone constraints, and resource boundary constraints. This indicates that all constraints are satisfied; This is an element-wise truncation operator used to preserve positive components that violate constraints; It is a norm used to aggregate the violations of various constraints into a single scalar for use in consistency judgment and threshold triggering.

[0037] In this embodiment, to ensure that the impact of constraint state variables on the project execution boundary can be incorporated into subsequent consistency judgment and consistency constraint training using a unified index, this embodiment defines a feasibility score to characterize the credibility of the state vector within the constraint domain, and establishes a monotonic correspondence between it and the constraint violation degree: in, State vector The feasibility score ranges from 0 to 1 and decreases as the degree of constraint violation increases; The degree of constraint violation; This is a positive coefficient used to adjust the sensitivity of the feasibility score to constraint violation. The value can be preset to a fixed value based on the management sensitivity of the evaluation cycle, so that the feasibility score can be used as a consistency judgment basis or as an input quantity for consistency constraints in subsequent model training and runtime verification.

[0038] Specifically, in step 300 of this embodiment, the project process data is first converted into a state sequence of the project twin state variable set according to the mapping relationship. This embodiment uses the evaluation period as the time division benchmark, dividing the project process data into multiple time intervals. The length of the evaluation period is consistent with the evaluation period determined in step 100. For example, when the evaluation period is 90 days, the project operation period is divided into multiple time intervals of consecutive 90 days; when the evaluation period is 180 days, the project operation period is divided into multiple time intervals of consecutive 180 days. For any given time interval, this embodiment, based on the mapping relationship, merges the project process data within that time interval into the periodic values ​​of each twin state variable in the project twin state variable set. The periodic values ​​are used to characterize the state level of the corresponding twin state variable within that time interval. For example, the periodic value of a schedule-related twin state variable can be determined by the correspondence between the number of completed tasks and the number of planned tasks within that time interval; the periodic value of a funding-related twin state variable can be determined by the correspondence between compliant expenditures and budgeted amounts within that time interval; and the periodic value of an outcome-related twin state variable can be determined by the number of outcomes that have passed acceptance within that time interval. Subsequently, this embodiment arranges the periodic values ​​of each time interval in chronological order to form a state sequence, enabling the state sequence to continuously depict the evolution trajectory of the project status as the evaluation cycle progresses.

[0039] After forming the state sequence, this embodiment performs consistency judgment on the state sequence under task dependency constraints, milestone constraints, and resource boundary constraints to obtain consistency judgment results. The consistency judgment here is used to determine whether the state values ​​of each evaluation period in the state sequence meet the execution boundaries defined by the project baseline data. Specifically, for each evaluation period, this embodiment judges whether the progress-related state values ​​meet the task dependency constraints defined by the task sequence relationship. For example, the completion status of a subsequent task in the current evaluation period cannot be determined as completed if the preceding task is not completed. It also judges whether the key state values ​​meet the milestone constraints defined by the milestone target and milestone date. For example, within an evaluation period including the milestone date, the result-related state values ​​should reach the number of results required by the milestone target. Finally, it judges whether the funding or resource-related state values ​​meet the resource boundary constraints defined by the funding limit or resource limit. For example, within a single evaluation period, the funding execution state values ​​cannot correspond to an expenditure level exceeding the budget limit. To ensure full disclosure, this embodiment may use no fewer than three milestone targets as the criteria for milestone constraints, and may limit the resource limit to no more than 160 hours of critical equipment usage time within the evaluation period or no more than the budget limit in the cumulative expenditure within the evaluation period, thereby giving the consistency criteria clear criteria and numerical boundaries.

[0040] When the consistency judgment result indicates that there are state values ​​in the state sequence that do not meet the task dependency constraint, milestone constraint, or resource boundary constraint, this embodiment corrects the state values ​​so that they meet the task dependency constraint, milestone constraint, and resource boundary constraint, and determines the corrected state sequence as the constraint-consistent feature sequence. Here, correction refers to adjusting the state values ​​that violate the constraints to a feasible range allowed by the constraints without changing the evaluation period division, so as to ensure that the feature sequence used for subsequent model training and prediction is interpretable and reproducible in terms of management constraints. For example, when a task dependency constraint indicates that a subsequent task must not be completed before a preceding task, and the state sequence shows a subsequent task completion state preceding a preceding task completion state, this embodiment adjusts the period value of the subsequent task completion state to be no higher than the period value of the preceding task completion state. When a milestone constraint requires at least three achievements within a certain evaluation period, but the state sequence shows insufficient results, this embodiment corrects the corresponding achievement state value to a lower limit consistent with the milestone target and marks it as a milestone constraint-driven correction. When a resource boundary constraint limits funding execution to not exceeding the budget limit, and the state value exceeds this limit, this embodiment corrects the funding state value to the upper limit corresponding to the budget limit. Through the above consistency judgment and correction, this embodiment obtains a constraint consistency feature sequence that satisfies the three types of constraints, providing stable input for subsequent consistency constraint training and runtime consistency verification.

[0041] After completing the consistency judgment and correction and determining the corrected state sequence as the constraint-consistent feature sequence, in order to make the constraint consistency process have a computable technical expression, this embodiment further expresses the constraint consistency correction of the state vector corresponding to each evaluation period as constraint projection: in, For the first The state vector of the project twin state variable set corresponding to each evaluation cycle. This is the constraint-consistent feature vector obtained after performing constraint consistency correction on the state vector. To constrain the projection operator, The feasible region is defined by task dependency constraints, milestone constraints, and resource boundary constraints. Let be the candidate vector within the feasible region. The L2 norm is used to measure the deviation between the candidate vector and the state vector, so that the deviation between the constraint-consistent feature vector and the state vector is minimized while satisfying the feasible region constraint, thereby achieving unified correction of state values ​​that violate task dependency constraints, milestone constraints, or resource boundary constraints.

[0042] In this embodiment, step 400 is used to train a performance prediction model based on a constrained consistent feature sequence, and to suppress violations of task dependency constraints, milestone constraints, and resource boundary constraints in the prediction output through consistency constraints. This embodiment uses the constrained consistent feature sequence obtained in step 300 as the input of the training samples, and uses performance labels obtained by converting historical performance review results according to the definitions, statistical granularity, aggregation rules, and evaluation cycles determined in step 100 as the supervision target. The performance labels are used to characterize the performance review conclusions corresponding to the evaluation cycle, and can be in the form of overall performance scores or sub-item performance scores. To ensure that the training samples cover the state changes of the project at different stages, this embodiment preferably selects no less than 12 evaluation cycles to form the training sample sequence, and ensures that the training samples cover no less than 2 milestone stages, thereby avoiding insufficient characterization of performance change patterns due to training only in a single stage.

[0043] During training, this embodiment drives the performance prediction model to output a performance prediction result within a prediction window for each feature sequence in the training samples. The prediction window is used to limit the future time range of the model output and corresponds to the evaluation period. For example, the prediction window can be one evaluation period or two consecutive evaluation periods. The performance prediction result can be the overall performance score sequence within the prediction window or the sub-item performance score sequence within the prediction window. This embodiment preferably sets the prediction window to one evaluation period to match the rolling prediction scenario, and uses no less than three types of sub-item performance labels in the training samples to cover dimensions such as progress, funding, and results, so that the model can learn the influence relationship of feature sequences on different performance dimensions.

[0044] After obtaining the performance prediction results, this embodiment substitutes these results into task dependency constraints, milestone constraints, and resource boundary constraints for consistency evaluation to obtain task dependency inconsistency, milestone inconsistency, and resource boundary inconsistency. Task dependency inconsistency measures whether the progress status implied by the performance prediction results violates the task sequence; milestone inconsistency measures whether the stage achievement level corresponding to the performance prediction results meets the requirements of the milestone goals and milestone dates; and resource boundary inconsistency measures whether the funding or resource consumption level corresponding to the performance prediction results exceeds the funding or resource limits. Subsequently, this embodiment determines the consistency constraint penalty based on the task dependency inconsistency, milestone inconsistency, and resource boundary inconsistency. This penalty is used to penalize prediction outputs that violate constraints, making the training process tend to output performance prediction results that satisfy the three types of constraints. To ensure that the consistency evaluation has repeatable quantitative boundaries, this embodiment preferably sets the lower limit of the number of milestone achievement items to 3, and sets the upper limit of the key equipment occupancy time in the resource boundary constraints to 160 hours or limits the cumulative expenditure within the evaluation period to no more than the budget limit, thus providing clear numerical conditions for the inconsistency judgment criteria.

[0045] This embodiment further determines the prediction error based on the performance prediction results and performance labels. The prediction error is used to measure the deviation between the predicted output and the historical performance review conclusions. The training objective is determined by combining the prediction error with the consistency constraint penalty, ensuring that the training objective simultaneously reflects prediction accuracy and constraint consistency. This embodiment trains the parameters of the performance prediction model under the constraints of the training objective, reducing both the consistency constraint penalty and the prediction error. This suppresses inconsistencies between the performance prediction results and task dependency constraints, milestone constraints, and resource boundary constraints, and improves the fit of the performance prediction results to historical performance review conclusions. To further ensure training stability, this embodiment preferably determines training convergence when both the prediction error and the consistency constraint penalty decrease simultaneously within five consecutive training rounds. After training, the model is validated using historical data from at least three evaluation periods to confirm that it can maintain consistent prediction outputs even in evaluation periods where it was not trained.

[0046] In determining the prediction error, this example quantifies the deviation between the performance prediction result output by the performance prediction model within the prediction window and the corresponding performance label as the prediction error, specifically as follows: in, This is the amount of prediction error; The number of evaluation periods included in the prediction window; For the performance prediction model in the first The performance prediction result vector output for each evaluation cycle; The performance label vector is obtained by converting historical performance review results according to the definition, statistical granularity, aggregation rules and evaluation period; The norm 2 is used to measure the deviation between the performance prediction result vector and the performance label vector. In determining the consistency constraint penalty and forming the training objective, this example quantifies the degree of violation of task dependency constraints, milestone constraints, and resource boundary constraints by the prediction output as the consistency constraint penalty, and uses the prediction error and the consistency constraint penalty together to determine the training objective, specifically: in, The penalty amount for consistency constraints; For the first The predicted state vector corresponding to each evaluation cycle; Let the constraint function vector consist of task dependency constraints, milestone constraints, and resource boundary constraints, and satisfy the following conditions: This indicates that all constraints are satisfied; For element-wise truncation operators; It is a norm; For training objectives; The coefficient is a positive number and remains fixed during training, used to balance the prediction error and the consistency constraint penalty.

[0047] In this embodiment, step 500 occurs during the project operation period. This embodiment acquires newly added project process data at the end of each evaluation cycle or when a preset update point is reached. The newly added project process data includes progress records, budget execution records, achievement registration records, or risk event records that are newly formed or collected relative to the previous evaluation cycle. Preferably, this embodiment uses the evaluation cycle as the update rhythm, ensuring that the newly added project process data covers the most recent evaluation cycle. For example, if the evaluation cycle is 90 days, new project process data is generated every 90 days. Alternatively, shorter update points can be set within the evaluation cycle to improve prediction timeliness, such as generating new project process data every 30 days and updating it on a rolling basis within the same evaluation cycle. Through this method, the newly added project process data can continuously reflect the status changes during project operation, providing a continuous data source for subsequent updates to the project digital twin.

[0048] After acquiring new project process data, this embodiment converts the new project process data into update values ​​for the project digital twin state variable set based on the aforementioned mapping relationship. The project digital twin is then updated within the feasible boundaries defined by task dependency constraints, milestone constraints, and resource boundary constraints, ensuring that the project digital twin maintains a state expression consistent with the actual process during runtime. For example, when the new project process data indicates that 5 tasks were completed within the last 30 days and the planned number of tasks is 8, this embodiment updates the progress-related digital twin state variables; when the new project process data indicates that 200,000 in compliant expenditures were added within the last 30 days and the corresponding budget amount is 500,000, this embodiment updates the funding-related digital twin state variables; when the new project process data indicates that 3 new accepted deliverables were added, this embodiment updates the deliverable-related digital twin state variables. If the updated state value reaches the resource limit or is inconsistent with the milestone date requirements, this embodiment ensures that the update result satisfies the resource boundary constraints and milestone constraints, ensuring that the state evolution of the project digital twin is always within the allowable range of management constraints, thereby avoiding runtime state drift that could cause distortion in the prediction output.

[0049] After the project digital twin is updated, this embodiment generates a corresponding feature sequence based on the updated project digital twin. This feature sequence characterizes the project's state evolution information over multiple consecutive evaluation periods, maintaining the same caliber and dimension as the feature sequence used in the training phase. Preferably, this embodiment sets the backtracking length of the feature sequence to four consecutive evaluation periods, ensuring the feature sequence covers at least one milestone stage. With an evaluation period of 90 days, the backtracking length of four evaluation periods corresponds to 360 days of state evolution information. If the project is in its early stages, resulting in fewer than four available evaluation periods, this embodiment uses existing evaluation periods to form the feature sequence, ensuring the feature sequence includes at least two evaluation periods to guarantee that the prediction input contains basic time-varying information. Through these limitations, the feature sequence maintains sensitivity to recent changes while also covering stage-specific evolutionary features.

[0050] After obtaining the updated feature sequence, this embodiment inputs the updated feature sequence into the performance prediction model to output the performance prediction results and confidence levels within a preset prediction window. The preset prediction window is used to limit the time span of future performance predictions. Preferably, this embodiment sets the preset prediction window to one evaluation period or two consecutive evaluation periods. For example, when the evaluation period is 90 days, the preset prediction window corresponds to the performance prediction results for the next 90 days or the next 180 days. The confidence level is used to characterize the reliability of the performance prediction results. This embodiment can limit the confidence level to a value between 0 and 1, and determine the confidence level based on the stability of the output distribution formed by the performance prediction model during the training phase. For example, when the fluctuation of the prediction results output by the performance prediction model for the same feature sequence is less than 0.05, the confidence level is determined to be no less than 0.80; when the fluctuation of the prediction results is greater than 0.10, the confidence level is determined to be no more than 0.60. By outputting the performance prediction results and confidence levels, this embodiment can provide rolling performance trend predictions with confidence representation during project operation, providing a basis for subsequent calibration and consistency verification.

[0051] When outputting the performance prediction results, this embodiment specifies the output of the performance prediction model to the updated feature sequence as the performance prediction result vector for each evaluation period within the prediction window, and arranges the performance prediction result vectors within the prediction window into a performance prediction result sequence in chronological order, specifically as follows: in, For the first Each evaluation period is a sequence of performance prediction results corresponding to the current point in time; For the first A vector of performance prediction results for each evaluation period; This refers to the number of evaluation periods included in the prediction window. In determining the confidence level, this embodiment defines the confidence level as the stability of the predicted output under small perturbations. Multiple predicted outputs are obtained by applying a constrained perturbation to the updated feature sequence, and the confidence level is then determined based on the dispersion of these multiple predicted outputs. Specifically: in, For the first Confidence level corresponding to each evaluation period; The first constrained perturbation applied to the updated feature sequence to obtain the... A vector of performance prediction results for each evaluation period; The second constrained perturbation applied to the updated feature sequence is used to obtain the... A vector of performance prediction results for each evaluation period; The norm is 2. In determining the constrained perturbation, this embodiment limits the constrained perturbation to a perturbation of the feature sequence that does not violate task dependency constraints, milestone constraints, and resource boundary constraints, and limits the perturbation amplitude to a proportional perturbation matching the magnitude of the feature values, specifically: in, For the first The evaluation period corresponding to the first Group of constrained perturbation feature vectors; For the first Constrained consistent feature vectors corresponding to each evaluation period; This is element-wise multiplication; It is a vector consisting entirely of 1s; For the first Group perturbation vector.

[0052] In this embodiment, step 600 is used to perform quality gating on the performance prediction results and confidence levels output in step 500 during runtime. A calibration action is triggered when the confidence level falls below a threshold or the constraint consistency check fails, in order to maintain the stability and interpretability of dynamic performance prediction. The confidence threshold is used to limit the minimum level of confidence of the prediction results. In this embodiment, the confidence threshold can be set to 0.70, and threshold judgment is performed immediately after the performance prediction output is completed in each evaluation cycle. When the confidence level is not lower than 0.70, this embodiment further performs a constraint consistency check on the performance prediction results to ensure that the prediction results meet the feasible boundaries defined by task dependency constraints, milestone constraints, and resource boundary constraints, thereby avoiding prediction outputs with high confidence levels but violating management constraints.

[0053] In performing constraint consistency checks, this embodiment performs task dependency consistency checks, milestone consistency checks, and resource boundary consistency checks on the performance forecast results. The task dependency consistency check determines whether the progress status corresponding to the performance forecast results satisfies the task sequence relationship defined by the task dependency constraint. For example, if the task breakdown structure stipulates that a subsequent task cannot be considered complete until a preceding task is completed, this embodiment determines that the completion status of a subsequent task within the forecast window should not precede the completion status of a preceding task. The milestone consistency check determines whether the stage outcome status corresponding to the performance forecast results meets the milestone achievement conditions defined by the milestone constraint. For example, if the milestone date falls within the forecast window, the predicted value of the number of outcomes should not be lower than the lower limit of the outcome specified in the milestone target. This embodiment can set this lower limit of outcomes to three items as the discrimination boundary. The resource boundary consistency check determines whether the funding or resource consumption status corresponding to the performance forecast results meets the upper limit of the resource boundary constraint. For example, this embodiment determines that the predicted value of cumulative expenditure within the evaluation period should not exceed the budget upper limit, or that the predicted value of the occupancy time of key equipment should not exceed 160 hours. By performing these three types of checks separately, the constraint consistency checks have clear discrimination objects and repeatable numerical boundaries.

[0054] When any of the task dependency consistency test, milestone consistency test, and resource boundary consistency test fails to meet the corresponding constraints, this embodiment determines that the constraint consistency test has failed and triggers an update of the mapping relationship or an adjustment of the performance prediction model parameters. Specifically, when the failure manifests as a shift in the scope of project process data fields or a change in the aggregation boundary leading to a systematic deviation in the performance prediction results, this embodiment triggers an update of the mapping relationship, ensuring that the correspondence between the performance indicator system and the project process data fields once again meets the definitions, statistical granularity, aggregation rules, and evaluation periods. When the failure manifests as a continuously widening deviation between the predicted output and the performance label, or a continuously lower confidence level, while maintaining a consistent mapping relationship, this embodiment triggers an adjustment of the performance prediction model parameters, allowing the performance prediction model to readjust to the changes in the state distribution during the project's operation. To ensure the triggering conditions are executable, this embodiment can limit "continuous" to two consecutive evaluation periods where the confidence level is below 0.70 or two consecutive evaluation periods experience any consistency test failure, thus giving the triggering criteria a clear time scale.

[0055] When the task dependency consistency test, milestone consistency test, and resource boundary consistency test all meet their respective constraints and the confidence level is not lower than the threshold, this embodiment determines that the constraint consistency test has passed and uses the current mapping relationship and the current performance prediction model to perform subsequent dynamic performance prediction. After triggering an update of the mapping relationship or adjusting the parameters of the performance prediction model, this embodiment uses the updated mapping relationship or the adjusted performance prediction model to regenerate the feature sequence and output the performance prediction results and confidence levels within the subsequent prediction window to verify the improvement effect of the calibration action on constraint consistency and prediction stability. To improve the traceability of the calibration results, this embodiment can review the results of the three types of consistency tests within a subsequent evaluation cycle, and maintain the updated mapping relationship or the adjusted performance prediction model for continuous operation when all consistency tests pass and the confidence level recovers to above 0.70.

[0056] When performing the constraint consistency test, to ensure that the three types of consistency tests have a unified computable expression, this embodiment summarizes the results of the task dependency consistency test, milestone consistency test, and resource boundary consistency test into the constraint violation degree within the prediction window, and expresses whether the constraint consistency test passes as a threshold determination of the constraint violation degree, specifically as follows: in, For the first When the evaluation period is the current point in time, the constraint violation degree of the performance prediction results within the prediction window; when When the task dependency consistency check, the milestone consistency check, and the resource boundary consistency check all satisfy their corresponding constraints, the constraint consistency check is considered passed. This indicates that at least one type of consistency check fails to meet the corresponding constraint, thus determining that the constraint consistency check has failed. When quantifying the triggering condition where the confidence level is below the threshold, this embodiment expresses the triggering criterion as a joint gating of confidence level and constraint violation, thereby avoiding false triggering caused by using only a single indicator. Specifically: in, For the first The trigger criterion value for each evaluation period and This indicates that the mapping relationship has been updated or the parameters of the performance prediction model have been adjusted. For indicator functions; The confidence level output in step 500; The confidence threshold; To constrain the violation threshold; This is a logical OR operation, used to represent that triggering occurs when any condition is met. When triggering the adjustment of the performance prediction model's parameters, to ensure that the "adjusted" result has a verifiable convergence basis, this embodiment describes the adjustment objective as simultaneously reducing the prediction error and the consistency constraint penalty, and uses the decrease in the training target as the criterion for determining the effectiveness of the adjustment. Specifically: in, For the first The decrease in the training objective before and after the adjustment is triggered in each evaluation cycle; The pre-adjustment training target is calculated based on the current feature sequence and the current performance label when the adjustment is triggered; To achieve the adjusted training objective calculated based on the same feature sequence and the same performance label after parameter adjustment; when This indicates that the adjustment lowers the training objective and serves as the basis for using the adjusted performance prediction model to perform subsequent dynamic performance predictions.

[0057] Corresponding to the above methods, such as Figure 2 As shown, this embodiment also provides a dynamic prediction system for the performance of science and technology projects that combines digital twins and artificial intelligence, including: The data mapping and indicator definition unit is used to acquire the performance indicator system and project process data of the target science and technology project, and to determine the definition, statistical granularity, collection rules and evaluation cycle of each indicator in the performance indicator system, and to establish the mapping relationship between the performance indicator system and the project process data fields. The twin construction and constraint setting unit is used to define the set of project twin state variables based on the performance indicator system and project baseline data, and to set task dependency constraints, milestone constraints and resource boundary constraints for the set of project twin state variables, so as to build a project digital twin that can be updated with project process data; The state sequence generation and feature extraction unit is used to convert project process data into a state sequence of project twin state variable sets according to the mapping relationship, and generate a constrained feature sequence under task dependency constraints, milestone constraints and resource boundary constraints. The consistency constraint model training unit is used to train the performance prediction model with feature sequences as input and performance labels obtained by transforming historical performance review results according to definition, statistical granularity, aggregation rules and evaluation cycle as output. It introduces consistency constraints to suppress inconsistencies between the prediction output and task dependency constraints, milestone constraints and resource boundary constraints. The dynamic update and performance prediction unit is used to update the project digital twin and update the feature sequence based on the newly added project process data during the project operation period. The updated feature sequence is input into the performance prediction model and the performance prediction results and confidence levels are output within the preset prediction window. The model adaptive correction and optimization unit is used to trigger the updating of the mapping relationship or the adjustment of the parameters of the performance prediction model when the confidence level is lower than the threshold or the constraint consistency test performed on the performance prediction results based on task dependency constraints, milestone constraints and resource boundary constraints fails. The updated mapping relationship or the adjusted performance prediction model is then used to perform subsequent dynamic performance prediction.

[0058] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0059] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for dynamically predicting the performance of science and technology projects by combining digital twins and artificial intelligence, characterized in that, include: Obtain the performance indicator system and project process data of the target technology project, and determine the definition, statistical granularity, collection rules and evaluation cycle of each indicator in the performance indicator system, and establish the mapping relationship between the performance indicator system and the project process data fields; Based on the performance indicator system and project baseline data, a set of project twin state variables is defined, and task dependency constraints, milestone constraints, and resource boundary constraints are set for the set of project twin state variables to construct a project digital twin that can be updated with the project process data; The project process data is converted into a state sequence of the project twin state variable set according to the mapping relationship, and a constraint-consistent feature sequence is generated under the task dependency constraint, the milestone constraint, and the resource boundary constraint; The performance prediction model is trained using the feature sequence as input and the performance labels obtained by converting historical performance review results according to the definition, statistical granularity, aggregation rules and evaluation cycle as output. Consistency constraints are introduced to suppress inconsistencies between the prediction output and the task dependency constraints, milestone constraints and resource boundary constraints. During the project operation period, the project digital twin is updated based on the newly added project process data, and the feature sequence is updated. The updated feature sequence is input into the performance prediction model to output the performance prediction results and confidence level within the preset prediction window. When the confidence level is lower than the threshold or the constraint consistency check performed on the performance prediction result based on the task dependency constraint, the milestone constraint and the resource boundary constraint fails, the mapping relationship is updated or the parameters of the performance prediction model are adjusted, and subsequent dynamic performance prediction is performed using the updated mapping relationship or the adjusted performance prediction model.

2. The method for dynamic prediction of science and technology project performance combining digital twins and artificial intelligence according to claim 1, characterized in that, Acquire the performance indicator system and project process data of the target technology project, and determine the definition, statistical granularity, collection rules, and evaluation cycle of each indicator in the performance indicator system. Establish the mapping relationship between the performance indicator system and the project process data fields, including: Obtain the indicator description information that records the definition, statistical granularity, aggregation rules and evaluation cycle of each indicator in the performance indicator system; Based on the indicator description information, the definition, statistical granularity, aggregation rules, and evaluation period of each indicator in the performance indicator system are determined respectively; Based on the aggregation rules, the project process data fields corresponding to each indicator in the performance indicator system are determined in the project process data, and the time aggregation method of the project process data fields is determined based on the statistical granularity and the evaluation period to form the mapping relationship.

3. The method for dynamic prediction of science and technology project performance combining digital twins and artificial intelligence according to claim 2, characterized in that, The process includes acquiring the performance indicator system and project process data of the target technology project, determining the definition, statistical granularity, aggregation rules, and evaluation cycle of each indicator in the performance indicator system, establishing the mapping relationship between the performance indicator system and the project process data fields, and further including: The definitions of each indicator in the performance indicator system are made consistent in terms of dimensions and units of measurement. The statistical granularity of each indicator in the performance indicator system is determined to be one of daily, weekly, or monthly. The evaluation period for each indicator in the performance indicator system is determined as a continuous time interval composed of multiple statistical granularities, and the evaluation period is aligned with the planned time boundary in the project baseline data.

4. The method for dynamic prediction of science and technology project performance combining digital twins and artificial intelligence according to claim 1, characterized in that, Based on the aforementioned performance indicator system and project baseline data, a set of project twin state variables is defined, including: Based on each indicator in the performance indicator system, define twin state variables that correspond one-to-one with each indicator; Based on the task decomposition structure, milestone plan, budget items and resource allocation in the project baseline data, define constraint state variables to characterize the task dependency constraints, the milestone constraints and the resource boundary constraints; The twin state variables and the constraint state variables are combined to form the project twin state variable set.

5. The method for dynamic prediction of science and technology project performance combining digital twins and artificial intelligence according to claim 4, characterized in that, Set task dependency constraints, milestone constraints, and resource boundary constraints for the project's twin state variable set, including: The task sequence relationship is determined based on the task decomposition structure in the project baseline data, and the task sequence relationship is used as the task dependency constraint. Based on the milestone plan in the project baseline data, milestone targets and milestone dates are determined, and the milestone targets and milestone dates are used as the milestone constraints. The funding cap or resource cap is determined based on the budget items and resource allocation in the project baseline data, and the funding cap or resource cap is used as the resource boundary constraint.

6. The method for dynamic prediction of science and technology project performance combining digital twins and artificial intelligence according to claim 1, characterized in that, The process data of the project is converted into a state sequence of the project twin state variable set according to the mapping relationship, including: The project process data is divided into multiple time intervals according to the evaluation cycle. For the project process data within each time interval, determine the periodic values ​​of each twin state variable in the project twin state variable set according to the mapping relationship; The period values ​​of each time interval are arranged in chronological order to obtain the state sequence.

7. The method for dynamic prediction of science and technology project performance combining digital twins and artificial intelligence according to claim 1, characterized in that, Generating a constrained feature sequence under the task dependency constraint, the milestone constraint, and the resource boundary constraint includes: Based on the task dependency constraints, the milestone constraints, and the resource boundary constraints, the consistency of the state sequence is determined to obtain the consistency determination result. The state values ​​in the state sequence that do not satisfy the task dependency constraint, the milestone constraint, or the resource boundary constraint are corrected so that the corrected state values ​​satisfy the task dependency constraint, the milestone constraint, and the resource boundary constraint. The corrected state sequence is determined as a characteristic sequence consistent with the constraints.

8. The method for dynamic prediction of science and technology project performance combining digital twins and artificial intelligence according to claim 1, characterized in that, A performance prediction model is trained using the aforementioned feature sequence as input and performance labels obtained from historical performance review results according to the aforementioned definition, statistical granularity, aggregation rules, and evaluation cycle as output. Consistency constraints are introduced to suppress inconsistencies between the prediction output and the aforementioned task dependency constraints, milestone constraints, and resource boundary constraints, including: Based on the feature sequences in the training samples, the performance prediction model outputs the performance prediction results within the prediction window; Substituting the performance prediction results into the task dependency constraint, the milestone constraint, and the resource boundary constraint respectively, we obtain the task dependency inconsistency, milestone inconsistency, and resource boundary inconsistency. The consistency constraint penalty is determined based on the task dependency inconsistency, the milestone inconsistency, and the resource boundary inconsistency. The prediction error is determined based on the performance prediction results and the performance labels. The training objective is determined by the prediction error and the consistency constraint penalty, and the parameters of the performance prediction model are trained according to the training objective to reduce the consistency constraint penalty and the prediction error, thereby suppressing the inconsistency between the performance prediction result and the task dependency constraint, the milestone constraint and the resource boundary constraint.

9. The method for dynamic prediction of science and technology project performance combining digital twins and artificial intelligence according to claim 1, characterized in that, When the confidence level is below a threshold or the constraint consistency check performed on the performance prediction result based on the task dependency constraint, the milestone constraint, and the resource boundary constraint fails, the mapping relationship is updated or the parameters of the performance prediction model are adjusted, including: The performance prediction results are subjected to task dependency consistency test, milestone consistency test, and resource boundary consistency test, respectively. The task dependency consistency test is used to determine whether the performance prediction results meet the task sequence relationship defined by the task dependency constraint; the milestone consistency test is used to determine whether the performance prediction results meet the milestone achievement conditions defined by the milestone constraint; and the resource boundary consistency test is used to determine whether the performance prediction results meet the resource upper limit conditions defined by the resource boundary constraint. If any of the task dependency consistency test, milestone consistency test, and resource boundary consistency test fails to meet the corresponding constraint, the constraint consistency test is determined to be unsuccessful, and the mapping relationship is updated or the parameters of the performance prediction model are adjusted. When the task dependency consistency check, the milestone consistency check, and the resource boundary consistency check all satisfy the corresponding constraints, the constraint consistency check is determined to be passed.

10. A dynamic performance prediction system for science and technology projects combining digital twins and artificial intelligence, characterized in that, include: The data mapping and indicator definition unit is used to acquire the performance indicator system and project process data of the target science and technology project, and to determine the definition, statistical granularity, collection rules and evaluation cycle of each indicator in the performance indicator system, and to establish the mapping relationship between the performance indicator system and the project process data fields. The twin construction and constraint setting unit is used to define a set of project twin state variables based on the performance indicator system and project baseline data, and to set task dependency constraints, milestone constraints and resource boundary constraints for the set of project twin state variables, so as to construct a project digital twin that can be updated with the project process data; The state sequence generation and feature extraction unit is used to convert the project process data into a state sequence of the project twin state variable set according to the mapping relationship, and generate a constrained feature sequence under the task dependency constraint, the milestone constraint and the resource boundary constraint. The consistency constraint model training unit is used to train a performance prediction model with the feature sequence as input and the performance labels obtained by converting historical performance review results according to the definition, the statistical granularity, the aggregation rule and the evaluation period as output. It introduces consistency constraints to suppress inconsistencies between the prediction output and the task dependency constraint, the milestone constraint and the resource boundary constraint. The dynamic update and performance prediction unit is used to update the project digital twin and the feature sequence based on the newly added project process data during the project operation period, and input the updated feature sequence into the performance prediction model to output the performance prediction results and confidence level within the preset prediction window; The model adaptive correction and optimization unit is used to trigger the updating of the mapping relationship or the adjustment of the parameters of the performance prediction model when the confidence level is lower than the threshold or the constraint consistency test performed on the performance prediction result based on the task dependency constraint, the milestone constraint and the resource boundary constraint fails, and then use the updated mapping relationship or the adjusted performance prediction model to perform subsequent dynamic performance prediction.

Images