An artificial intelligence-based project collaborative schedule optimization management method and system

Abstract

This invention relates to the field of project management technology, and in particular to an artificial intelligence-based project collaborative progress optimization management method and system. The method includes: performing filtering and matching operations on all configured resources based on the core categories and hierarchical mapping relationships of project management to obtain a scenario-based configuration set; parsing project features and matching them with historical data using a feature matching algorithm to obtain a project and task association list; generating specific task time baselines using a progress prediction algorithm, and optimizing these baselines according to preset adjustment rules to obtain a final baseline; integrating task information and inspection results to form an identifiable task deliverable package; using a three-dimensional early warning model to issue graded early warnings for abnormal tasks, generating a monitoring list with early warning levels; reviewing abnormal tasks based on the monitoring list and providing feedback on the final results, integrating them to form an audit result set, and obtaining the calibrated progress. This solution effectively solves the problems of low efficiency and passive risk intervention in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of project management technology, and in particular to a project collaborative schedule optimization management method and system based on artificial intelligence. Background Technology

[0002] In modern enterprise operations, collaborative project schedule management is a crucial link in ensuring on-time project delivery, controlling operating costs, and improving the quality of deliverables. It is widely used in various fields such as software development, engineering construction, and product manufacturing. As project scale expands, cross-departmental collaboration increases, and personalized needs rise, the complexity of project management, including resource allocation, task assignment, schedule control, and anomaly intervention, increases significantly, placing higher demands on the precision, intelligence, and efficiency of management methods.

[0003] However, existing technologies have many shortcomings. Resource allocation and matching rely on manual processes, lack standardized hierarchical mapping and data verification mechanisms, resulting in poor adaptability and high redundancy. Furthermore, the intelligent analysis of project characteristics and solution generation is insufficient, failing to fully utilize historical data and experience, leading to highly subjective solutions with significant errors. Moreover, schedule baselines are often based on experience-based estimations, failing to comprehensively consider multiple factors, resulting in a disconnect from actual execution capabilities, and lacking quantifiable rules for adjustments to special needs. Anomaly warnings are triggered by a single dimension, the review and calibration process is cumbersome, collaborative management is inefficient, and risk intervention is reactive.

[0004] These problems result in insufficient adaptability, intelligence, and control efficiency in project management, making it difficult to meet the needs of modern enterprises for refined management. Therefore, there is an urgent need for a management method that can achieve precise resource matching, intelligent solution generation, scientific benchmark setting, and closed-loop control. Summary of the Invention

[0005] This invention establishes a standardized hierarchical mapping, intelligent feature matching, multi-dimensional progress prediction, graded early warning, and closed-loop audit and calibration mechanism to achieve precise matching of resource allocation, intelligent generation of task plans, scientific and controllable progress benchmarks, timely intervention of abnormal risks, and efficient closed-loop collaborative management. It effectively solves the problems of poor adaptability, insufficient intelligence, low management efficiency, and passive risk intervention of existing technologies.

[0006] The technical solution proposed in this invention is: a project collaborative progress optimization management method based on artificial intelligence, the method comprising: Based on the core categories and hierarchical mapping of project management, a filtering and matching operation is performed on all configuration resources, and data cleaning and compatibility verification are completed simultaneously to obtain a scenario-based configuration set. Based on the scenario-based configuration set, the feature matching algorithm is used to parse the project features and match them with historical data to generate task and resource plans and obtain a list of project and task associations. By combining the project and task association list, the reviewer's workload and historical data, a specific task time baseline is generated through a progress prediction algorithm. The specific task time baseline is then optimized according to preset adjustment rules to obtain the final baseline. Based on the final benchmark, advance the task and update the status, and integrate the task information and test results to form a labeled task outcome package; Based on the final progress benchmark and task deliverables, abnormal tasks are identified through a three-dimensional early warning model, and a monitoring list with warning levels is generated. Based on the monitoring list, abnormal tasks are reviewed and the final results are fed back. The results are integrated to form a review result set. At the same time, the task progress is dynamically adjusted based on the review comments to obtain the adjusted progress.

[0007] Preferably, the specific process for obtaining the scenario-based configuration set is as follows: A three-level hierarchical mapping system is constructed based on project management industry standards and practical data, and a rule library for each level is established. Add hierarchical tags to all configuration resources, label scene attributes, record resource mutual exclusion conditions, dependency order and priority, and form a configuration resource tag library; Using the project's core identifier as the primary key, we extract the associated mid-level functional categories from the configuration resource tag library, and remove irrelevant categories after scene attribute verification to obtain a valid mid-level category set. Based on the effective middle-level classification set, a candidate set of lower-level configuration elements is extracted. Combined with the association rule base, suitable resource combinations are screened, mutually exclusive resources and redundant elements are eliminated, and a preliminary configuration set is formed after hierarchical verification. The initial configuration set is categorized by function, deduplication is performed based on the core project identifier and time difference threshold, missing data is filled in by time-series interpolation, and abnormal data is detected and smoothed by sliding window to obtain noise-reduced configuration data. The noise reduction configuration data is supplemented with key configurations according to the core element list, inefficient and redundant resources are filtered out, and aggregated to form a scenario-based configuration set.

[0008] Preferably, the specific process for obtaining the project and task association list is as follows: Key features are extracted from the scenario-based configuration set. Categorical features are transformed using one-hot encoding, and continuous features are mapped through normalization to construct the project feature vector. Collect feature vectors, task decomposition structures, resource allocation details, and execution effect feedback from similar past projects. After cleaning and deduplication, classify and store them according to core dimensions to form a historical project knowledge base. By combining the analytic hierarchy process with gradient descent to optimize feature weights, cosine similarity to calculate dimensional fit, weighted summation to obtain comprehensive similarity, high-similarity historical projects to be screened and verified a second time, a set of effective reference cases is obtained. Extract common task logic and resource rules from a collection of effective reference cases, adjust task modules, resource specifications and execution order according to the current project scenario, add special requirement nodes, and form a preliminary task and resource plan; Clearly define the task leader, resource details, start and end times, and dependencies of the preliminary plan. After logical consistency verification, integrate them according to the preset template to form a project and task association list.

[0009] Preferably, the specific process for obtaining the final benchmark is as follows: Extract task types and review nodes from the associated list, collect reviewer load data, retrieve historical data of similar tasks, clean and remove abnormal data and standardize the format; The task characteristics, reviewer workload, and historical benchmarks are clearly defined, and the weight of each dimension is determined by expert scoring combined with historical data regression. Calculate the historical average time consumption by screening similar task sets, adjust it in combination with the current load saturation, and calibrate it according to personnel proficiency and delivery standards to obtain the initial baseline; Establish quantitative adjustment logic, verify process compliance, audit capacity, and task dependency coordination. If the verification is successful, determine the final benchmark; otherwise, optimize the adjustment ratio or coordinate the requirements.

[0010] Preferably, the specific process for obtaining the task outcome package is as follows: Synchronize the final baseline, task requirements, and dependencies to the executor, and generate a unique task execution code as the core index; The status is updated through time-triggered and operation feedback, and the status is confirmed and synchronized in real time to the executor and the manager. The implementing party uploads the deliverables, and the system automatically verifies the file format, content completeness, and key indicators, providing feedback on rectification items until the verification is passed; Using task execution coding as the core, assign basic identifiers, status identifiers, traceability identifiers, and audit-related identifiers to qualified deliverable packages; Integrate qualified deliverables documents and identifiers to obtain the task deliverable package.

[0011] Preferably, the specific process for obtaining the monitoring list is as follows: Extract the planned completion time and node requirements from the final baseline, synchronize the actual progress, verification results, and resource allocation data of the deliverables package, and integrate them into an early warning dataset; The schedule dimension calculates the deviation rate and overdue time; the quality dimension counts the number and severity of problems; the resource dimension calculates the gap rate and marks the gap type, and clarifies the triggering conditions and indicators for each level. Filter abnormal tasks and sort them in descending order of warning level. Integrate project name, abnormal type, core indicators, and rectification requirements to form a structured monitoring list.

[0012] Preferably, the specific process for obtaining the calibrated progress is as follows: Integrate monitoring lists, anomaly details, rectification feedback, and deliverables; The rectification plans and catch-up progress for abnormal progress are classified and verified; the deliverables and key issues for abnormal quality are reviewed; the resource replenishment plan for abnormal resources is confirmed; the rectification of each dimension of compound abnormalities is checked item by item; and the audit results are integrated to form a result set. If the review is approved, the baseline will be updated according to the actual progress or catch-up plan. If the review is partially approved, time will be reserved for supplementary rectification to adjust the baseline. If the review is rejected, the progress baseline will be reset. Archive the audit results set and calibration progress to the project data center, and associate the task code and exception record.

[0013] The present invention also provides an artificial intelligence-based project collaboration progress optimization management system, the system being used to execute the aforementioned artificial intelligence-based project collaboration progress optimization management method.

[0014] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the aforementioned artificial intelligence-based project collaborative progress optimization management method.

[0015] The beneficial effects of this invention are: 1. This solution constructs a three-level standardized hierarchical mapping system, combined with intelligent resource tagging preprocessing and multi-dimensional data cleaning and verification mechanisms, to achieve automatic filtering, matching and precise adaptation of configuration resources, effectively eliminate redundant resources and invalid data, significantly reduce manual screening costs, improve project startup efficiency, and solve the problems of poor resource adaptability and difficulty in ensuring data quality in existing technologies.

[0016] 2. By relying on feature matching algorithms to analyze the core features of the project and reusing historical project data experience, combined with multi-dimensional progress prediction and quantitative adjustment rules, the generated task and resource plans are more in line with the personalized needs of the project. The established progress benchmark is scientific and controllable and can flexibly respond to special requirements, avoiding the benchmark disconnect caused by experience estimation, and significantly improving the rationality of task allocation and the accuracy of progress control.

[0017] 3. By accurately identifying abnormal tasks and sorting them according to risk level through a three-dimensional hierarchical early warning model, combined with the systematic integration of pre-information system and closed-loop audit and calibration mechanism, timely intervention of abnormal risks, efficient advancement of audit process and dynamic optimization of progress data are achieved. This solves the defects of passive risk intervention, cumbersome audit and lagging progress data in existing technologies, and strengthens the systematicness and closed-loop nature of project collaborative management. Attached Figure Description

[0018] Figure 1A flowchart of an AI-based project collaborative schedule optimization management method; Figure 2 This is a flowchart illustrating the management process of an AI-based project collaborative progress optimization management method. Detailed Implementation

[0019] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0020] It is understood that the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0021] like Figure 1 and Figure 2 As shown, the system filters and matches core content according to hierarchical relationships to generate scenario-based configuration sets. Through hierarchical and precise filtering, irrelevant configuration options are eliminated, ensuring that the configuration content is highly adapted to the project scenario, thereby improving work efficiency. Based on the scenario-based configuration sets, a feature matching algorithm analyzes project characteristics and matches them with historical data to generate task and resource plans, obtaining a list of project and task associations. The algorithm automatically associates projects and tasks, reducing manual configuration costs and errors, while reusing past successful experiences to improve the rationality of task and resource allocation. Combining the project and task association list with the reviewer's workload and historical data, a progress prediction algorithm generates a specific task time baseline. If there are other requirements, the specific task time baseline is adjusted to obtain the final baseline. By comprehensively analyzing multi-dimensional key data, a scientific time reference is generated, making task time planning more aligned with actual execution capabilities and avoiding blindly setting progress targets. The executor advances the task according to the final baseline, and the system automatically updates the status, forming an identifiable task deliverable package. This clarifies the executor's work basis, enabling traceability of the task execution process, and ensures real-time synchronization of task progress data through automatic status updates. Based on the final schedule baseline and task deliverables, the system uses a 3D early warning model to issue tiered warnings for abnormal tasks, generating a monitoring list with warning levels. It identifies schedule risks in advance and pushes warning information precisely according to the level, facilitating managers to quickly locate problems and intervene in a timely manner, reducing the risk of project delays. Managers review the list, provide feedback on the final results, and obtain the review result set and calibrated schedule. Closed-loop management ensures that task deliverables meet quality standards, while dynamically calibrating project progress to keep schedule data accurate at all times.

[0022] Furthermore, the system filters and matches based on the core content categories according to hierarchical relationships to generate scenario-based configuration sets. The specific process is as follows: First, based on project management industry standards (such as the PMBOK Guidelines and the ISO 21500 system) and practical experience, a three-tiered mapping system is constructed, consisting of upper-level core categories, middle-level functional categories, and lower-level configuration elements. The upper-level core categories cover general classification dimensions such as project domain, management type, and implementation phase. The middle-level functional categories correspond to core project management modules such as task allocation, resource scheduling, and progress monitoring. The lower-level configuration elements include specific execution requirements such as templates, task pools, and resource lists. Simultaneously, a rule base for each level is established to clarify the many-to-many mapping relationship between upper-level categories and middle-level categories, the compatibility rules between middle-level categories and lower-level elements, and the dependency order. The rule base refers to a structured set of rules built based on historical project configuration data and industry standards, used to constrain the matching logic and data flow relationships between each level. In this process, the system introduces artificial intelligence technology, using the C4.5 decision tree algorithm to train on historical project configuration data from the past three years to build an association rule mining model: The first step uses project domain, management type, and implementation stage as input features of the decision tree, and the middle-level functional classification and the matching combination of lower-level configuration elements as the output target; the second step calculates the splitting priority of each feature using the information gain ratio (e.g., the information gain ratio of the project domain is 0.38, higher than the 0.25 of the management type, so it is prioritized as the root node); the third step iteratively splits the decision tree nodes, for example, under the project domain = automobile manufacturing branch, it further splits by implementation stage = final assembly, extracting the association rules between the final assembly stage and equipment scheduling and quality inspection; the fourth step prunes and optimizes the decision tree (removing branches with a depth exceeding 5 levels) to avoid overfitting. The final association rule library achieves a matching accuracy of 96%, and it continuously iterates with newly added project data each month, with a rule update cycle of no more than 7 days.

[0023] Next, all configuration resources undergo structured tagging preprocessing. Based on a three-level hierarchical mapping system, hierarchical tags are added to each configuration resource, with each tag corresponding one-to-one with the upper, middle, and lower layers. Scene attributes are labeled, including features such as the applicable project scale, execution mode, and industry segmentation. Relationships are recorded, including mutual exclusion conditions, dependency order, and priority rules between resources. This results in a structured configuration resource tag library containing resource IDs, hierarchical tags, scene attributes, and relationship rules. This tag library is a structured data collection used to store all configuration resources and their associated information, supporting rapid retrieval and matching. The system processes the configuration resource tag library using Word2Vec vector embedding technology from artificial intelligence: First, it converts hierarchical tags and scene attributes into text sequences (such as "Automobile Manufacturing - Final Assembly - Equipment List - Medium-sized Project"); second, it trains a word vector model using a sliding window with a window size of 5, mapping the text sequence of each resource into a 128-dimensional vector (each dimension of the vector takes values ​​in the range [-1,1]); finally, it constructs a vector index library. When searching for "New Energy Vehicles - Sedan - Mass Production", it quickly locates resources with a similarity ≥ 0.9 by calculating the cosine similarity (the dot product of vector A and vector B divided by the product of the magnitudes of the two vectors), with a retrieval response time ≤ 0.3 seconds, which is 4 times more efficient than traditional database retrieval.

[0024] Then, using the project's core category as the primary key, a hierarchical automated search and matching process is executed. First, based on the precise mapping relationship between the upper-level core category and hierarchical tags, a set of directly related mid-level functional categories is extracted from the configuration resource tag library. Category items unrelated to the current project scenario are eliminated through scenario attribute verification, resulting in a valid mid-level category set. Second, using the valid mid-level category set as the search basis, a corresponding candidate set of lower-level configuration elements is extracted. Combining compatibility rules and mutual exclusion conditions in the association rule library, suitable configuration resource combinations are selected, and mutually exclusive resources and redundant elements are eliminated. Third, hierarchical correlation verification is performed on the candidate configuration elements to ensure logical consistency between lower-level elements and the upper-level core category and mid-level functional categories, forming a preliminary configuration set. This process utilizes a graph neural network (GNN) model from artificial intelligence to perform hierarchical correlation verification: First, a correlation graph of "upper-level category - middle-level classification - lower-level element" is constructed, where each node contains attributes such as resource ID and label, and edges represent correlation relationships; second, the graph is trained using the GraphSAGE algorithm to learn the correlation weights between nodes (e.g., the correlation weight between "new energy vehicles" and "battery assembly template" is 0.92, and the correlation weight between "new energy vehicles" and "battery assembly template" is 0.15); finally, nodes with correlation weights < 0.6 (i.e., logically broken elements) are removed, increasing the logical consistency compliance rate of the initial configuration set from 88% to 99%.

[0025] Finally, the initial configuration set is optimized for completeness and adaptability. Based on a pre-set checklist of core elements, the initial configuration set is checked to see if it contains the critical configurations required for project startup. These critical configurations include basic templates, core task pools, and lists of necessary resources. For missing critical elements, they are automatically completed from the configuration resource tag library based on the association rule base. Duplicate and inefficient redundant configurations are filtered out. Finally, a scenario-based configuration set highly adapted to the project scenario is formed. Simultaneously, the system records the hierarchical association logs, resource adaptation rules, and verification results of this screening and matching process, providing data support for subsequent configuration system optimization. In this stage, artificial intelligence employs a dual-model approach for missing element detection and redundancy removal: Missing element detection utilizes a random forest algorithm (composed of 50 decision trees, with features including configuration element type and project scenario attributes). By training with complete configuration cases, the model can accurately identify 98% of missing elements (e.g., when the core equipment requirement list is missing, the model outputs a missing probability of 0.95) and selects the resource with the highest fit (≥90%) from the tag library to complete the missing elements; Redundancy removal uses the DBSCAN clustering algorithm, which clusters configuration resources according to functional similarity (with a cluster radius of 0.3). Only the latest version of the resource is retained in the same cluster, reducing the configuration set redundancy rate from 15% to below 3%.

[0026] Furthermore, based on the scenario-based configuration set, project features are analyzed using a feature matching algorithm and matched with historical data to generate task and resource plans, resulting in a list of associated projects and tasks. The specific process is as follows: First, core project features are extracted from the scenario-based configuration set to construct a standardized project feature vector. Based on the hierarchical labels and scenario attributes in the scenario-based configuration set, key feature dimensions such as project type, implementation scale, core objectives, time constraints, budget range, resource requirement type, and delivery standards are extracted. Each feature is quantified. Categorical features are converted into discrete values ​​using one-hot encoding, and continuous features are normalized to the [0,1] interval to ensure uniformity in the magnitude of features across different dimensions. Finally, a structured project feature vector is formed. This project feature vector refers to the transformation of multi-dimensional unstructured features of a project into a computable and comparable numerical vector, providing standardized input for subsequent similarity matching. The system introduces a multilayer perceptron (MLP) model from artificial intelligence to optimize feature vector construction: the model includes an input layer (12 feature dimensions), a hidden layer (2 layers, 32 neurons per layer, using ReLU activation function), and an output layer (12-dimensional vector, consistent with the input dimensions); by training on project feature data, the model can automatically learn the interaction relationships between features (e.g., the feature interaction weight between "new energy vehicles" and "battery technology" is increased to 0.2), improving the feature vector's ability to represent project requirements by 20%, and subsequently increasing the matching accuracy from 85% to 97%.

[0027] Secondly, a historical project data knowledge base is constructed. Complete execution data from similar past projects is collected, covering feature vectors, task decomposition structures, resource allocation details, process dependencies, and execution performance feedback (such as schedule achievement rate and resource utilization rate). Historical data undergoes standardized cleaning to remove invalid and redundant data, ensuring that its feature dimensions and data format are consistent with the current project's feature vector. A hierarchical indexing system is established based on feature dimensions, categorized and stored according to core dimensions such as project type and implementation scale, forming a historical project data knowledge base that supports rapid retrieval and accurate matching. The historical project data knowledge base refers to a structured data set that stores standardized historical project-related data to provide matching references for the current project. Artificial intelligence employs both data cleaning and semantic indexing techniques in this stage: data cleaning involves 3... Outliers are removed in principle (such as projects whose construction period exceeds the mean by ±3 standard deviations), and linear interpolation is used to fill in missing data within 10% (such as null values ​​for "resource utilization") to ensure data integrity of 99%. The semantic index is built on Elasticsearch and transforms project feature vectors into searchable text fields (such as "project type: new energy vehicles, scale: 500 units / month"), supports retrieval by feature combination (such as new energy vehicles and automated assembly), achieves a retrieval recall rate of 95%, and has a response time of ≤0.5 seconds.

[0028] Next, association matching is performed using a feature matching algorithm. The first step involves setting feature dimension weights. Based on the degree of impact of each feature on project execution (e.g., core objectives and time constraints have a more critical impact on project matching, thus having a higher weight than budget range and resource requirement type), a corresponding weight coefficient is assigned to each feature dimension. Key features have higher weight coefficients to ensure they dominate the matching results. Artificial intelligence optimizes the weights using the Analytic Hierarchy Process (AHP) combined with gradient descent: First, 10 project management experts (with ≥8 years of experience) are invited to construct a judgment matrix to assess the importance of "core objectives" and "budget range." Pairwise comparisons (using a 1-9 scale, where 1 indicates equal importance and 9 indicates extreme importance) yielded initial weights. Next, using historical project matching accuracy as the objective function, the weights were optimized using a gradient descent algorithm (learning rate 0.01, 100 iterations). The final weights were determined as follows: core objective weight 0.18, time constraint 0.15, project type 0.12, implementation scale 0.10, delivery standard 0.09, budget range 0.08, resource requirements 0.07, and other features totaling 0.21. The weight consistency test CR = 0.06 < 0.1, meeting the reasonableness requirements.

[0029] The second step is to perform dimensional similarity comparison. Using the current project feature vector as a benchmark, the feature vectors of each historical project in the historical project data knowledge base are extracted one by one. The quantified values ​​of corresponding dimensions in the two sets of vectors are compared one by one. If the difference between the two sets of values ​​is smaller, the degree of fit for that dimension is higher, and vice versa. The system achieves dimensional comparison through the cosine similarity algorithm. Specifically, the dot product of the current vector and the historical vector is calculated by multiplying the value of the first dimension of the current vector with the value of the first dimension of the historical vector, multiplying the value of the second dimension with the value of the second dimension, and so on, until the 12th dimension is multiplied. The 12 products are then added together to obtain the dot product. Next, the magnitude of the current vector is calculated by squaring the value of each dimension of the current vector, summing them, and taking the arithmetic square root. The magnitude of the historical vector is calculated in the same way. Finally, the dot product is divided by the product of the magnitudes of the two vectors to obtain the cosine similarity (range 0-1). For example, a similarity of 0.92 indicates that the two vectors have a very high degree of fit, while 0.5 indicates a low degree of fit.

[0030] The third step is to calculate the overall similarity score. First, multiply the fit of each dimension by its corresponding weight coefficient. Then, sum the products of all dimensions. The final sum is the overall similarity score between the current project and the historical project. A higher score indicates a stronger fit in core features. For example, the similarity scores of the current project and a historical project are as follows: core objective 0.95, time constraint 0.90, project type 0.98, with corresponding weights of 0.18, 0.15, and 0.12. The average similarity of other dimensions is 0.85, and the total weight is 0.55. Therefore, the overall similarity score is = That is, 89.11 points.

[0031] The fourth step is to perform threshold screening. Based on industry practice and project management experience, a fixed similarity threshold (such as 0.8) is set. Historical projects with a comprehensive similarity score higher than this threshold are selected to form a high-matching historical project case set. This threshold is verified through 500 sets of historical project data: when the threshold is set to 0.8, there are 460 cases that are successfully matched (i.e., the historical solution is adapted to the current project), with a success rate of 92%; the success rate is 88% when the threshold is 0.75, and 90% when the threshold is 0.85. Therefore, 0.8 is the optimal threshold, which can balance the success rate and the number of cases (case set size ≥ 20 sets).

[0032] The fifth step involves conducting a secondary verification process. This involves checking the compatibility and effectiveness of historical projects in the case set, focusing on whether there are any conflicts between their resource allocation models and the current project's scenario-based configuration, and whether the execution results meet the preset excellent standards. Cases with conflicts and inefficient cases with poor execution results are removed to obtain a set of effective reference cases. The secondary verification is completed by a logistic regression model: the model uses the number of resource allocation conflicts, the project schedule achievement rate, and the resource utilization rate as features, and effectiveness as a label. Through training with historical cases, the model achieves an accuracy rate of 94%. For example, if a historical case has two resource allocation conflicts with the current project and a project schedule achievement rate of 85%, the model outputs an invalid probability of 0.91, and it is removed.

[0033] Then, tasks and resource plans are generated based on the current project scenario. The system extracts common task decomposition logic, resource allocation rules, and inter-task dependencies from a set of effective reference cases. Combined with the current project's scenario-based configuration set, it adjusts task module division, resource specification matching, and task execution order to suit specific project needs (such as time constraints and budget limitations). It clarifies the core objectives, execution standards, required resource types and quantities, and estimated time consumption for each task, forming a preliminary task and resource plan. Artificial intelligence optimizes the plan generation through transfer learning algorithms: using a Transformer-based transfer learning model, the system uses task plans from the effective reference case set as source domain data (500+ plans) and the current project scenario as target domain data. The model first learns the mapping relationship between task types and resource requirements in the source domain (e.g., a final assembly task corresponds to 2 final assembly lines and 5 technicians). Then, through domain adaptation techniques (e.g., adversarial training), it transfers knowledge from the source domain to the target domain, automatically deleting tasks unnecessary in the target domain (e.g., "off-road kit assembly" specific to SUVs) and adding special tasks in the target domain (e.g., hybrid battery installation). Finally, it outputs a preliminary plan, improving adjustment efficiency by 5 times compared to manual adjustments, and achieving a ≥92% adaptability to the current scenario.

[0034] Finally, a project and task association list is compiled. The initial task and resource plans are structured and clearly defined, specifying the specific tasks, responsible persons, required resources, start and end times, and pre- and post-task relationships for each project module. Logical consistency checks ensure no omissions in task decomposition, no conflicts in resource allocation, and no contradictions in relationships. The compiled content is then integrated according to preset field templates to generate a project and task association list containing three-dimensional association information of projects, tasks, and resources. Simultaneously, the system records the source of historical cases, similarity scores, and plan adjustment records for this matching, providing data support for subsequent plan optimization. Artificial intelligence utilizes both topological sorting and knowledge graph technologies for sorting and verification: Topological sorting resolves task dependency conflicts by constructing a directed graph of tasks based on dependencies, using the Kahn algorithm (prioritizing tasks with in-degree 0) to ensure no circular dependencies, achieving 100% sorting accuracy; the knowledge graph is used for resource allocation verification, constructing a graph of resources, time, and tasks. Conflicts are identified by querying whether the same resource is allocated to multiple tasks within the same time period, achieving a conflict identification rate of 98% and ensuring a 99% logical consistency rate for the association list.

[0035] Furthermore, by combining the project and task association list with the reviewer's workload and historical data, a specific task time baseline is generated using a progress prediction algorithm. If there are other requirements, the specific task time baseline is adjusted to obtain the final baseline. The specific process is as follows: First, we collected and standardized the input data. We extracted core information from the project and task association list, including task type, task size, number of review nodes, delivery standards, and dependencies between tasks. We collected reviewer load data, covering the total number of tasks currently being handled by the reviewer, the staffing of the review team, the average daily review volume per person, and load saturation (i.e., the ratio of current task volume to maximum review capacity). Reviewer load saturation is a core indicator of the tightness of review resources, with a value range of [0,1]. The closer the value is to 1, the tighter the review resources are. We retrieved historical review data, including the actual review time for similar tasks in the past, the reviewer load during the corresponding period, adjustment records during the review process, and final pass rate feedback. We cleaned and deduplicated the historical data, removing abnormal time-consuming data caused by sudden failures or missing materials to ensure data validity and consistency. Artificial intelligence is deployed in this stage as a data preprocessing pipeline: First, calculate the load saturation: Load saturation = Total number of tasks currently being handled by the reviewer ÷ (Review team staffing × Average daily review volume per person × Review cycle days), where the review cycle days are 30 days by default and can be adjusted according to project needs (e.g., 15 days for urgent projects). For example, if there are currently 12 tasks being handled, 8 staff members, an average of 2 tasks per person per day, and a cycle of 30 days, then the load saturation = 12 ÷ (8 × 2 × 30) = 12 ÷ 480 = 0.025; Second, historical data cleaning uses the box plot method: First, calculate the two key quartiles Q1 (25th percentile, i.e., all data are sorted from...) of historical time-consuming data. After sorting the data from smallest to largest, the values ​​at the 25th percentile and Q3 (75th percentile, i.e., the values ​​at the 75th percentile after sorting all data from smallest to largest) are calculated. Then, the interquartile range is calculated, which is the difference between Q3 and Q1. Next, two boundaries for outliers are determined: the lower limit is Q1 minus 1.5 times the interquartile range, and the upper limit is Q3 plus 1.5 times the interquartile range. Finally, extreme data exceeding these upper and lower boundaries are removed (such as 50 hours of data consumption due to equipment failure, which far exceeds the boundary of 30 hours), so that the data validity reaches 95%. The third step is to unify the data format: convert the time units such as "minutes" and "hours" to "hours" and retain one decimal place to ensure data format consistency.

[0036] Secondly, a core influencing factor system for progress forecasting is constructed. Three core influencing dimensions are identified: task characteristics (task type, scale, number of review nodes), reviewer load status (load saturation, reviewer proficiency), and historical execution benchmarks (average historical review time for similar tasks, reasonable fluctuation range). Weight coefficients are assigned based on the degree of influence of each dimension on review time. Reviewer load saturation and historical review time for similar tasks have the highest weight, followed by the number of task review nodes. For example, the weight of reviewer load saturation is set to 0.4, historical review time for similar tasks to 0.3, reviewer proficiency to 0.15, the number of task review nodes to 0.1, and the strictness of task delivery standards to 0.05, ensuring that key factors play a dominant role in forecasting. The weighting coefficients are determined through expert scoring and historical data regression: For expert scoring, eight senior experts (≥10 years of experience) in the review field are invited to score the importance of each dimension (1-10 points), and the average score is used to obtain the expert weight; for historical data regression, a multiple linear regression model is used, with actual review time as the dependent variable and data for each dimension as independent variables, and data weights are obtained by training historical samples; the final weight = expert weight × 0.6 + data weight × 0.4. For example, the expert weight for load saturation is 0.38, and the data weight is 0.42, so the final weight = 0.38 × 0.6 + 0.42 × 0.4 = 0.228 + 0.168 = 0.396 ≈ 0.4, which meets the preset requirements. Data across all dimensions is quantified. Task delivery standards are categorized into three levels based on industry norms: Level 1 (standard) requires only one initial review and complete documentation, with no additional verification required. Level 2 (stricter standard) requires an initial review and one round of cross-verification, with supporting documentation. Level 3 (stricter standard) requires three rounds of review: initial review, cross-verification, and final review, with detailed process documentation and third-party verification reports required. Reviewer proficiency is assessed based on both years of experience and the pass rate over the past six months. The indicators are categorized by coefficients ranging from 0.7 to 1.0. 0.7 (entry-level) corresponds to less than 1 year of experience and an 85%-90% pass rate; 0.8 (basic proficiency level) corresponds to 1-3 years of experience and a 91%-95% pass rate; 0.9 (proficient professional level) corresponds to 3-5 years of experience and a 96%-98% pass rate; and 1.0 (senior expert level) corresponds to ≥5 years of experience and a pass rate ≥99%. Higher coefficients indicate higher review efficiency and lower error rates. This coefficient classification is based on historical data statistics of reviewers: Pearson correlation analysis shows a strong positive correlation between years of experience and pass rate (0.82), and a strong negative correlation between pass rate and review time (-0.79). For example, a reviewer with 3 years of experience and a 97% pass rate (coefficient 0.9) will have approximately 15% less review time than a reviewer with a coefficient of 0.7, closely matching actual data.

[0037] Next, the initial task time baseline is calculated using a progress prediction algorithm. Using task type and the number of review nodes as search criteria, the most similar task sets are selected from standardized historical data. The average review time, median review time, and reasonable fluctuation range (e.g., ±10%) for this set are calculated. Dynamic adjustments are made based on the current reviewer load saturation. If the reviewer load saturation is ≥0.8 (high load), the time is increased by 15%-35% based on the historical average (the higher the load, the greater the increase); if the load saturation is ≤0.3 (low load), the time is decreased by 8%-18%. Further calibration is performed based on the reviewer proficiency coefficient and the strictness of the task delivery standards. For example, a reviewer proficiency of 0.7 increases the time by 10%-20% compared to 1.0, and a Level 3 delivery standard increases the time by 5%-15% compared to Level 1. Combining these adjustments, an initial review time baseline for each specific task is generated, along with the calculation basis (e.g., referenced historical cases, load adjustment ratios, etc.). The progress prediction algorithm uses a textual step-by-step calculation: First, filter similar task sets: for example, using engine quality inspection and 4 audit nodes as conditions, 30 sets of historical tasks are selected, with an average time of 8 hours, a median of 7.8 hours, and a fluctuation range of 7.2-8.8 hours; Second, load adjustment: the current load saturation is 0.85 (high load), adjusted according to a piecewise function: when the load is 0.8-0.9, it increases by 25%, so the adjusted time = 8 × (1 + 25%) = 10 hours; Third, proficiency and standard calibration: audit... The nuclear personnel proficiency level is 0.8 (basic proficiency), which is 15% higher than 1.0. The Level 3 standard is 12% higher than Level 1. Therefore, the calibration time is 10 × (1 + 15%) × (1 + 12%) = 10 × 1.15 × 1.12 = 12.88 hours, which means the initial baseline is 12.9 hours (rounded to one decimal place). The fourth step is to mark the calculation basis: refer to "Engine Quality Inspection Task (Group 30) from October to December 2023", with a load adjustment of 25%, a proficiency level of 0.8 calibration of 15%, and a Level 3 standard calibration of 12%.

[0038] Next, clarify the types of special requirements and adjustment rules. Identify common special requirements, including accelerating the review process (e.g., shortening the review cycle due to earlier project milestones), increasing review resources (e.g., temporarily adding reviewers), adjusting the review scope (e.g., adding / reducing some review content), and raising review standards (e.g., adding a review stage). Develop quantitative adjustment rules for each requirement. For example, for requirements to accelerate the review process, calculate the maximum reasonable compression space using the formula "compressible ratio = 1 - (current load saturation × 0.6)," and the compressed time must not be lower than the historical minimum necessary review time for similar tasks (i.e., the historical shortest review time after excluding outliers). For increasing review resources, each additional skilled reviewer can reduce the time by 8%-12% from the initial baseline (cumulative reduction not exceeding a 30% cap). For adjusting the review scope, adjust the time proportionally based on the proportion of newly added / reduced review content to the original task; adding a review stage will additionally increase the time by 10%-15% from the original baseline. Example calculations for each rule: Current load saturation is 0.7, compressible ratio = 1 - (0.7 × 0.6) = 1 - 0.42 = 0.58, initial baseline is 12.9 hours, compressed time = 12.9 × (1 - 0.58) = 5.418 hours, historical minimum necessary time is 6 hours, so we take 6 hours; Add 2 skilled personnel (skill level 1.0), each person's time is reduced by 10%, cumulative reduction is 20%, adjusted time = 12.9 × (1 - 20%) = 10.3 hours (not exceeding the 30% limit); Add 20% more review content, adjusted time = 12.9 × (1 + 20%) = 15.5 hours; Upgrade from level 2 to level 3, increase by 12%, adjusted time = 12.9 × (1 + 12%) = 14.4 hours.

[0039] Finally, the final benchmark is determined and verified. Special requirements from users are received, and the initial task time benchmark is adjusted according to the corresponding adjustment rules to obtain an adjusted candidate benchmark. The candidate benchmark is then verified for feasibility, checking whether it meets three conditions: "compliance of the review process (e.g., necessary review steps cannot be reduced)," "actual capacity of the reviewer (e.g., the adjusted time is not less than the minimum review time for a single person per task)," and "coordination between tasks (e.g., no conflict with the delivery time of preceding tasks or the start time of subsequent tasks)." If the verification passes, it is determined as the final benchmark. If the verification fails, the process returns to the adjustment stage to optimize the proportion, or the reasonableness of the special requirements is communicated and coordinated with the user until a feasible final benchmark is formed. The system synchronously records the initial benchmark, the basis for adjustment, the content of special requirements, and the generation log of the final benchmark, providing data support for subsequent progress tracking and algorithm optimization. Feasibility verification was performed using a 3D verification model: The verification checked whether the Level 3 standard retained the three rounds of preliminary review, cross-review, and final review, with each round taking a minimum of 2 hours. The candidate baseline of 6 hours ≥ 3 × 2 = 6 hours, indicating compliance. The audit team consisted of 8 people, and the adjusted time was 6 hours. The average workload per person was 6 ÷ (8 × 8) = 6 ÷ 64 ≈ 0.094 ≤ 1.2 (within a reasonable range), indicating feasibility. The delivery time for the pre-task was 10.1, and the start time for the post-task was 10.8. The candidate baseline of 6 hours corresponded to a cycle of 10.1-10.3, with no conflict. Verification passed, and the final baseline was 6 hours. If the candidate baseline was 5 hours (< 6 hours), the process compliance verification failed, and the process was returned to the adjustment stage to reduce the compressible ratio to 0.5. The compressed time was 12.9 × (1 - 0.5) = 6.45 hours ≈ 6.5 hours, requiring re-verification.

[0040] Furthermore, the executor advances the task based on the final baseline, and the system automatically updates the status, generating a task deliverable package with identification tags. The specific process is as follows: First, the system synchronizes front-end information and displays core indexes before task execution. It synchronizes the final task time baseline, core task requirements (including delivery standards, review nodes, and resource configuration details), and inter-task dependencies to the executor's front-end interface. A unique task execution code is generated and displayed on the front-end interface. This code serves as the core index for front-end status tracking, results uploading, and review association, and is permanently displayed at the top of the executor's interface throughout the entire task execution process. Artificial intelligence generates and optimizes the task execution code at this stage: the code uses a structure of project feature code, timestamp, and random sequence. The project feature code consists of the first letter of the project domain, the first letter of the type, and the scale level (e.g., "NE-EQ-M" corresponds to "New Energy Vehicles - Engine Quality Inspection - Medium"). The timestamp is a 10-digit second-level time (e.g., 1717200000), and the random sequence is a 4-digit number (e.g., 1234). An example code is "NE-EQ-M-1717200000-1234". Simultaneously, SHA-256 encryption is used to ensure uniqueness (repetition rate < 1234). Furthermore, the code can be quickly associated with task information through AI semantic parsing (entering the code will return "New Energy Vehicle Engine Quality Inspection Task, Medium Scale, Created on June 1, 2024").

[0041] Next, the executor proceeds with the task according to the baseline, and the front end displays the status update results in real time. Based on the final time baseline, the executor formulates a phased execution plan on the front end and advances the work according to the task nodes; the system updates the task status through a dual mechanism of automatic triggering at time nodes and feedback from the executor's front end operation, and displays the status indicator in a prominent position on the front end interface. The task status visible on the front end is divided into 5 categories: Pending (marking the planned start time), In Progress (displaying the current node's completion percentage), Paused (the front end pop-up displays the reason for the pause, such as resource shortage, requirement change, and the pause application review result), Overdue Warning (marking the overdue duration and overdue percentage), and Completed (displaying the actual completion time); after all status updates, they are synchronized in real time to the interface shared by the executor and the manager to ensure that the information on both ends is consistent. The system incorporates a real-time status management algorithm based on artificial intelligence: Status updates utilize the WebSocket protocol for bidirectional communication. After the executor submits an operation (such as "pause application"), the front end sends the task ID, operation type, and time to the back end via JSON format. After the back end updates the status database, it pushes the status change to all associated terminals (executor and manager) through a message broadcast mechanism, with a delay of ≤1 second. Status anomaly identification employs a time window algorithm: A 5-minute monitoring window is set. If a "in progress" task shows no progress updates within the window, the AI ​​automatically triggers a front-end pop-up prompt (such as "Engine quality inspection task has not shown any progress updates for 5 minutes, please check the execution status"). The anomaly identification rate reaches 98%, avoiding status lag.

[0042] Next, the front-end deliverables are uploaded and compliance verification feedback is conducted. After completing the task, the implementing party uploads complete deliverables (including process data, final outputs, and execution summary) through the front-end. The front-end interface displays the file upload progress, format prompts, and verification of required fields. The system automatically extracts the core information of the deliverables and compares it with the delivery standards. The verification results are fed back to the front-end in real time: if the verification passes, "Verification Passes" is displayed; if there are problems such as format errors or missing content, the front-end interface lists specific rectification items (such as "Missing execution process record sheet" or "File format is not the specified PDF version"). The implementing party makes supplementary modifications on the front-end and re-uploads for verification until it passes. This stage deploys a multi-dimensional intelligent verification pipeline: It uses file magic number recognition (e.g., PDF magic number is "%PDF-1.7", Word magic number is "D0CF11E0") to determine the format with 100% accuracy. If a JPG file (not a specified format) is uploaded, it prompts "Please upload a PDF / Word file"; it uses OCR technology (recognition rate ≥99%) to extract the file text, combined with a TF-IDF keyword matching algorithm (keywords include "execution record", "acceptance conclusion", "test data"); if the matching degree is <80%, it is judged as missing content. For example, if the "execution process record table" is missing, it prompts "Please supplement the key steps record in the execution process"; it extracts key indicators from the file (e.g., "pass rate 98%" "number of tests 5") and compares them with the delivery standard (e.g., pass rate ≥95%). If the standard is not met, it prompts "Pass rate does not meet 95% standard, please retest"; the verification pass rate is improved from 85% for manual verification to 96%.

[0043] Then, assign values ​​to the front-end display identifiers of qualified deliverables. The system uses the task execution code as the core and displays multi-dimensional identifiers for qualified deliverable packages on the front end, specifically including: a basic identifier area (displaying the task execution code, project name, executor name, and task start / completion time), a status identifier area (displaying the execution status, verification results, and benchmark compliance, with benchmark compliance displayed as a percentage and marked "Meets benchmark / Requires special explanation"), a traceability identifier area (displaying front-end jump links for status change records, resource usage details, and pause / rectification records), and an audit association identifier area (displaying the corresponding auditor name, preset audit node, and audit deadline). Automatically calculate and populate identification information: Calculate the baseline fit, baseline fit = (final baseline time ÷ actual time) × 100%. For example, if the baseline is 6 hours and the actual time is 5.8 hours, the fit = (6 ÷ 5.8) × 100% ≈ 103.4%; if the actual time is 7 hours, the fit = (6 ÷ 7) × 100% ≈ 85.7%; calculate based on the actual completion time and 2 working days. For example, if the completion time is October 1st (Wednesday), the deadline = October 1st + 2 = October 3rd (Friday); if the completion time is October 4th (Saturday), the deadline = October 4th + 2 = October 7th (Monday); generate a redirect URL and associate it with the corresponding record in the status log table (e.g., "Status Change Record Link" points to the "Status History Page of Task ID=123") to ensure that clicking it allows you to view details.

[0044] Finally, the front-end integrates and archives the identified task deliverables into a unified "task deliverable package." The system integrates verified deliverable files with the aforementioned identification information into a single package. The front-end interface supports online previewing, downloading, and forwarding of the deliverable package to the reviewer. The deliverable package is simultaneously archived on the front-end, categorized and searchable by task execution code. Clicking on a code allows users to view the entire task execution process data (baseline information, status update logs, and verification records). The system records the deliverable package generation time, identification assignment details, and verification records on the front-end, facilitating traceability and verification by the executor and administrator.

[0045] Furthermore, based on the final progress baseline and task deliverables, the system uses a three-dimensional early warning model to issue graded early warnings for abnormal tasks, generating a monitoring list with early warning levels. The specific process is as follows: First, the system integrates and synchronizes the input data with the front end. It extracts core data from the final progress baseline (planned task completion time, progress requirements for each node, and time baseline), synchronizes real-time execution data from the task deliverables package (actual completed progress, time elapsed, details of uploaded deliverables), resource allocation data from the executor (types and quantities of resources in place, resource utilization rate), and deliverable verification data (format verification results, content completeness verification records, and key indicator compliance status). This data is then integrated into a standardized dataset, displayed on the front end as a task overview (total number of tasks, number of normal tasks, number of abnormal tasks) and a core data overview. Data presentation is optimized through data fusion and visualization technologies: a weighted average method is used to fill in missing data. For example, when resource utilization rates are empty, the weighted average of the resource utilization rate of similar tasks (weight 0.7) and the average utilization rate of the reviewers (weight 0.3) is used, achieving 99% data completeness.

[0046] Secondly, clarify the dimensions of three-dimensional anomaly identification and the quantitative display on the front end. Progress deviation dimension: Using the final progress benchmark as a reference, calculate the progress completion deviation rate ((actual completed progress - planned concurrent progress) × 100%) and the time consumption deviation rate. The front end directly displays the overdue duration (overdue items are highlighted in the common screenshot style). Deliverables quality dimension: Based on the verification results of the task deliverables package, the number of verification issues and their severity (general / important / critical) are counted, and the number of issues and their severity indicators are displayed on the front end, and a pop-up window for details of the issues is linked simultaneously; Resource matching dimension: Compare resource demand with actual availability, calculate resource gap rate, and display availability status on the front end with a visible resource configuration progress bar, mark the types of missing resources, and simultaneously display the review status of resource supplementation applications.

[0047] Next, establish rules for identifying abnormal tasks: a task is considered abnormal and added to the monitoring list if it meets any of the following conditions: Schedule dimension anomalies: Schedule completion deviation rate ≤-5% (lagging) or ≥15% (excessive ahead of schedule), or there is overdue time; Quality dimension anomalies: ≥2 verification issues, or including important / critical level issues; Resource dimension anomaly: resource gap rate ≥10%, or the missing resources are the core resources required for the task; Composite anomalies: Anomalies exist simultaneously in two or more of the above dimensions.

[0048] Then, the system filters out abnormal tasks and generates a front-end monitoring list. The system automatically filters out abnormal tasks according to the above rules, integrates them into a structured monitoring list, and sorts them by priority from compound anomalies to single-dimensional anomalies. Filtering can be done by anomaly type and executor, and an anomaly details pop-up window can be expanded.

[0049] Ultimately, the system enables front-end push and linkage of abnormal information. It synchronizes complex abnormalities and abnormal tasks containing critical issues to managers and implementers via in-site messages and pop-up notifications. Implementers can upload rectification progress, and the system updates task status in real time. It records the abnormality identification time, viewing status, rectification feedback records, and status change trajectory, forming a full lifecycle tracking log for abnormal tasks, which is then synchronously archived in the project data center.

[0050] Furthermore, managers review the checklist, provide feedback on the final results, and obtain the review result set and the progress after calibration. The specific process is as follows: First, the system integrates and displays pre-audit information on the front end. It integrates information such as the abnormal task monitoring list, three-dimensional abnormal details, rectification feedback records from the executor, task deliverables (including supplementary documents after rectification), and historical audit data into the front end. The content includes project name, task name, abnormality type, core abnormality information, rectification feedback content, rectification attachments (supporting online preview), executor, and current status. Managers can quickly locate tasks to be audited using filtering controls (abnormality type, executor) and view the complete abnormality and rectification trajectory. Employing an attention mechanism algorithm, the system prioritizes historical similar audit cases and before-and-after rectification comparison data associated with the current audit task. For example, when auditing engine quality inspection abnormalities, it automatically displays audit opinions on similar quality issues from the past three months to assist managers in decision-making. The system supports natural language retrieval; for example, inputting "new energy vehicle battery audit" allows the front end to match tasks using NLP word segmentation (keywords "new energy vehicle," "battery," and "audit"), achieving a 92% accuracy rate, a 30% improvement over traditional keyword retrieval.

[0051] Secondly, the administrator performs front-end review. The administrator reviews the complete information of each abnormal task and performs the corresponding review actions for different abnormal types: Progress anomalies: Verify the feasibility of the rectification plan, whether the implementation party's promised catch-up progress is reasonable, and compare the actual progress after rectification with the planned progress; Quality anomalies: Review the rectified deliverables, verify whether key issues have been resolved, and check the results of the system's secondary verification. Resource anomaly: Confirm whether the missing resources have been replenished or there is a clear replenishment plan, and check the progress of resource arrival; Complex anomalies: Check the rectification status of each dimension item by item to ensure no issues are missed; supports three types of operations: approval, rejection, and partial approval. After clicking, a screenshot-adapted opinion input box will pop up, and you need to fill in the clear review opinions before submitting.

[0052] Next, a structured audit result set is generated. The system automatically integrates the audit results based on the manager's audit operations to form a standardized audit result set. The core fields include: task execution code, project name, task name, exception type, audit result (pass / reject / partially pass), audit comments (fully retaining the manager's input), auditor, audit time, rectified items, and non-rectified items. The audit result set is displayed in reverse chronological order on the front-end module and supports filtering by audit result, exception type, and auditor. The executor can view the corresponding results through the front-end.

[0053] Then, the task progress is calibrated based on the audit results. If the audit is approved: if the progress meets the requirements after rectification, the system will use the current actual progress as the calibrated progress; if there is a reasonable catch-up plan, the calibrated progress baseline will be updated according to the plan, and the front end will synchronously display the calibrated planned completion time. Partially approved: For items that did not meet the standards, the progress benchmark was adjusted based on the audit comments, a reasonable time was reserved for rectification and supplementation, and the deadline for supplementary rectification was displayed on the front end; Review rejection: The progress benchmark is reset according to the review comments, and the progress is calibrated with the re-review after rectification is completed. The front end highlights the prompt that re-rectification is required and the core reasons for rejection; the calibrated progress is synchronized to the implementer and the manager in real time.

[0054] Finally, the results are archived and updated in conjunction with the front end. The audit result set and the calibrated progress are archived synchronously to the project data center and stored in association with the task execution code, exception records, and rectification logs; the front end automatically updates the task status: if the audit is passed, the exception status is removed; if partially passed / rejected, the exception mark is retained, and the rectification requirements and deadlines are updated synchronously; the system records the entire audit process log (including audit operation time, audit comments, and progress calibration basis).

[0055] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0056] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0057] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. A project collaborative progress optimization management method based on artificial intelligence, characterized in that, The method includes: Based on the core categories and hierarchical mapping of project management, a filtering and matching operation is performed on all configuration resources, and data cleaning and compatibility verification are completed simultaneously to obtain a scenario-based configuration set. Based on the scenario-based configuration set, the feature matching algorithm is used to analyze the project features and match historical data to generate task and resource plans and obtain a list of project and task associations. By combining the project and task association list, the reviewer's workload and historical data, a specific task time baseline is generated through a progress prediction algorithm. The specific task time baseline is then optimized according to preset adjustment rules to obtain the final baseline. Based on the final benchmark, advance the task and update the status, and integrate the task information and test results to form a labeled task outcome package; Based on the final progress benchmark and task deliverables, abnormal tasks are identified through a three-dimensional early warning model, and a monitoring list with warning levels is generated. Based on the monitoring list, abnormal tasks are reviewed and the final results are fed back. The results are integrated to form a review result set. At the same time, the task progress is dynamically adjusted based on the review comments to obtain the adjusted progress.

2. The project collaborative progress optimization management method based on artificial intelligence according to claim 1, characterized in that, The specific process for obtaining the scenario-based configuration set is as follows: A three-level hierarchical mapping system is constructed based on project management industry standards and practical data, and a rule library for each level is established. Add hierarchical tags to all configuration resources, label scene attributes, record resource mutual exclusion conditions, dependency order and priority, and form a configuration resource tag library; Using the project's core identifier as the primary key, we extract the associated mid-level functional categories from the configuration resource tag library, and remove irrelevant categories after scene attribute verification to obtain a valid mid-level category set. Based on the effective middle-level classification set, a candidate set of lower-level configuration elements is extracted. Combined with the association rule base, suitable resource combinations are screened, mutually exclusive resources and redundant elements are eliminated, and a preliminary configuration set is formed after hierarchical verification. The initial configuration set is categorized by function, deduplication is performed based on the core project identifier and time difference threshold, missing data is filled in by time-series interpolation, and abnormal data is detected and smoothed by sliding window to obtain noise-reduced configuration data. The noise reduction configuration data is supplemented with key configurations according to the core element list, inefficient and redundant resources are filtered out, and aggregated to form a scenario-based configuration set.

3. The project collaborative progress optimization management method based on artificial intelligence according to claim 2, characterized in that, The specific process for obtaining the project and task association list is as follows: Key features are extracted from the scenario-based configuration set. Categorical features are transformed using one-hot encoding, and continuous features are mapped through normalization to construct the project feature vector. Collect feature vectors, task decomposition structures, resource allocation details, and execution effect feedback from similar past projects. After cleaning and deduplication, classify and store them according to core dimensions to form a historical project knowledge base. By combining the analytic hierarchy process with gradient descent to optimize feature weights, cosine similarity to calculate dimensional fit, weighted summation to obtain comprehensive similarity, and screening of historical projects with high similarity for secondary verification, a set of effective reference cases is obtained. Extract common task logic and resource rules from a collection of effective reference cases, adjust task modules, resource specifications and execution order according to the current project scenario, add special requirement nodes, and form a preliminary task and resource plan; Clearly define the task leader, resource details, start and end times, and dependencies of the preliminary plan. After logical consistency verification, integrate them according to the preset template to form a project and task association list.

4. The project collaborative progress optimization management method based on artificial intelligence according to claim 3, characterized in that, The specific process for obtaining the final benchmark is as follows: Extract task types and review nodes from the associated list, collect reviewer load data, retrieve historical data of similar tasks, clean and remove abnormal data and standardize the format; The task characteristics, reviewer workload, and historical benchmarks are clearly defined, and the weight of each dimension is determined by expert scoring combined with historical data regression. Calculate the historical average time consumption by screening similar task sets, adjust it in combination with the current load saturation, and calibrate it according to personnel proficiency and delivery standards to obtain the initial baseline; Establish quantitative adjustment logic, verify process compliance, audit capacity, and task dependency coordination. If the verification is successful, determine the final benchmark; otherwise, optimize the adjustment ratio or coordinate the requirements.

5. The project collaborative progress optimization management method based on artificial intelligence according to claim 4, characterized in that, The specific process for obtaining the task outcome package is as follows: Synchronize the final baseline, task requirements, and dependencies to the executor, and generate a unique task execution code as the core index; The status is updated through time-triggered and operation feedback, and the status is confirmed and synchronized in real time to the executor and the manager. The implementing party uploads the deliverables, and the system automatically verifies the file format, content completeness, and key indicators, providing feedback on rectification items until the verification is passed; Using task execution coding as the core, assign basic identifiers, status identifiers, traceability identifiers, and audit-related identifiers to qualified deliverable packages; Integrate qualified deliverables documents and identifiers to obtain the task deliverable package.

6. The project collaborative progress optimization management method based on artificial intelligence according to claim 5, characterized in that, The specific process for obtaining the monitoring list is as follows: Extract the planned completion time and node requirements from the final baseline, synchronize the actual progress, verification results, and resource allocation data of the deliverables package, and integrate them into an early warning dataset; The schedule dimension calculates the deviation rate and overdue time; the quality dimension counts the number and severity of problems; the resource dimension calculates the gap rate and marks the gap type, and clarifies the triggering conditions and indicators for each level. Filter abnormal tasks and sort them in descending order of warning level. Integrate project name, abnormal type, core indicators, and rectification requirements to form a structured monitoring list.

7. The project collaborative progress optimization management method based on artificial intelligence according to claim 6, characterized in that, The specific process for obtaining the calibrated progress is as follows: Integrate monitoring lists, anomaly details, rectification feedback, and deliverables; The rectification plans and catch-up progress for abnormal progress are classified and verified; the deliverables and key issues for abnormal quality are reviewed; the resource replenishment plan for abnormal resources is confirmed; the rectification of each dimension of compound abnormalities is checked item by item; and the audit results are integrated to form a result set. If the review is approved, the baseline will be updated according to the actual progress or catch-up plan. If the review is partially approved, time will be reserved for supplementary rectification to adjust the baseline. If the review is rejected, the progress baseline will be reset. Archive the audit results set and calibration progress to the project data center, and associate the task code and exception record.

8. A project collaborative progress optimization management system based on artificial intelligence, characterized in that, The system is used to execute the project collaborative progress optimization management method based on artificial intelligence as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the artificial intelligence-based project collaborative progress optimization management method according to any one of claims 1-7.

Images