Digital project management method and system based on artificial intelligence

Abstract

The invention discloses a digital project management method and system based on artificial intelligence, and relates to the technical field of project management, and the method comprises the steps: constructing a project digital portrait and multi-dimensional feature analysis framework, determining an analysis domain boundary of intelligent auditing, and automatically generating an intelligent analysis grid adaptive to multi-source heterogeneous data; and performing grid division on project progress data according to an analysis domain boundary, performing adaptive partition on a project grid through an unsupervised clustering algorithm, dynamically determining a critical threshold value of risk early warning based on a project digital portrait and intelligent analysis grid and comprehensive risk stress analysis, and outputting a risk early warning decision management conclusion containing the critical threshold value. According to the management system, an intelligent analysis grid adaptive to multi-source heterogeneous data is generated by constructing a project digital portrait and a multi-dimensional feature analysis framework, and a risk early warning critical threshold value is dynamically determined, so that real-time perception, accurate analysis and intelligent decision support of the full life cycle state of a project are realized.

Description

Technical Field

[0001] This invention relates to the field of project management technology, and specifically to a digital project management method and system based on artificial intelligence. Background Technology

[0002] In modern engineering project management practice, the entire project lifecycle encompasses multiple stages, including planning, execution, monitoring, and closure. Its management effectiveness directly impacts project delivery quality, resource utilization efficiency, and risk control levels. However, traditional project management models primarily rely on human experience, which has limitations and fails to achieve accurate characterization and dynamic adaptation of project status.

[0003] Traditional technologies face multiple challenges, especially in project schedule control and material characteristic analysis: First, project progress data is usually recorded in the form of discrete nodes or segmented milestones, lacking fine-grained gridded representation, making it difficult to identify key progress bottlenecks or abnormal fluctuations through adaptive partitioning. Secondly, the characterization of material characteristics (such as specifications, usage distribution, supply chain stability, etc.) relies on manual statistics or static templates, which cannot dynamically capture the micro-change patterns during the material's use. Third, risk warnings rely on preset fixed thresholds (such as the number of days of project delay or the percentage of cost overruns), without being dynamically adjusted in conjunction with digital profiling of the project and multi-dimensional feature analysis. This results in a significant lag in warnings, and by the time actual risks accumulate to the trigger threshold, they have often already caused irreversible damage to the overall project schedule or quality.

[0004] Given the special requirements of engineering projects for efficient collaboration, precise control, and proactive risk prevention, it is necessary to propose an artificial intelligence-based digital project management method. This method involves constructing a digital profile of the project and a multi-dimensional feature analysis framework to generate an intelligent analysis grid that adapts to multi-source heterogeneous data and dynamically determining the critical threshold for risk warning. This enables real-time perception, precise analysis, and intelligent decision support for the entire lifecycle status of the project. Summary of the Invention

[0005] The purpose of this invention is to provide a digital project management method and system based on artificial intelligence to address the shortcomings in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a digital project management method based on artificial intelligence, the management method comprising the following steps: S1. Based on the needs of project lifecycle management, the project entity is digitally abstracted, a digital profile of the project and a multi-dimensional feature analysis framework are constructed, and the analysis domain boundary of intelligent audit is established; S2. Based on the digital profile of the project, an intelligent analysis grid is automatically generated to adapt to multi-source heterogeneous data. The intelligent analysis grid includes a project progress grid and a material feature characterization grid. The project progress data is divided into grids according to the boundary of the analysis domain. The project grid is adaptively partitioned by an unsupervised clustering algorithm, and the grid density parameters are dynamically adjusted for each partition. S3. Based on the project's digital profile and intelligent analysis grid, comprehensively analyze the risk stress, dynamically determine the critical threshold for risk warning, and output risk warning decision management conclusions that include the critical threshold.

[0007] In a preferred embodiment, step S3 involves dynamically determining the critical threshold for risk warning based on the project's digital profile and intelligent analysis grid, combined with comprehensive risk stress analysis, including the following steps: The equivalent scenario of simulating node advancement is used to compare the project progress update process to dynamic influencing factors acting on control points. By calculating the response intensity of material declaration data when the project status changes, the risk amplitude distribution is generated. Based on the analysis grid and risk interaction model, the risk amplitude of each monitoring point is calculated, and a risk distribution cloud map is synthesized. The flow field structure of the risk distribution cloud map is then analyzed. Identify risk clustering areas, risk gradient abrupt change zones, and risk vortex regions. Based on the flow field structure analysis results, map the risk amplitude to project management risk factors and calculate the intensity and direction of each risk factor. By comprehensively analyzing the risk forces, the critical threshold for risk warning is dynamically determined.

[0008] In a preferred embodiment, the output of a risk warning decision management conclusion containing a critical threshold includes the following steps: In terms of progress management, if the risk magnitude of a certain milestone node exceeds the dynamic critical threshold, an early warning will be triggered.

[0009] In a preferred embodiment, the output of a risk warning decision management conclusion containing a critical threshold includes the following steps: For the materials management dimension, if the risk amplitude of a certain element in the material characteristic characterization grid exceeds the dynamic critical threshold, it is marked as a risk missing item.

[0010] In a preferred embodiment, step S2 involves automatically generating an intelligent analysis grid adapted to multi-source heterogeneous data based on the project's digital profile, including the following steps: The project schedule data is divided into grids based on the boundary conditions of the analysis domain. The project timeline is discretized into several time step units, and a three-dimensional grid is generated for each schedule node, represented as: time dimension × task dimension × resource dimension.

[0011] In a preferred embodiment, step S2 involves automatically generating an intelligent analysis grid adapted to multi-source heterogeneous data based on the project's digital profile, including the following steps: To address the multimodal characteristics of material profiling, a feature representation grid is constructed for each material type. For textual materials, a concept space grid is generated through semantic embedding vectors; for tabular materials, a structured grid is generated through numerical field mapping; and for image / video materials, a visual grid is generated through computer vision feature extraction.

[0012] In a preferred embodiment, step S2 involves dividing the project progress data into grids based on the analysis domain boundaries, adaptively partitioning the project grid using an unsupervised clustering algorithm, and dynamically adjusting the grid density parameters for each partition, including the following steps: A composite analysis grid system is formed by integrating the dynamic observation sequences of N monitoring indicators with the baseline grid of P control observation points; The project grid is adaptively partitioned using an unsupervised clustering algorithm. The grid density parameter is dynamically adjusted for each partition, and the monitoring weight of the grid nodes is optimized based on the issue distribution in the historical audit case library.

[0013] In a preferred embodiment, step S1 involves digitally abstracting the project entity based on the project's full lifecycle management requirements, constructing a digital profile and multi-dimensional feature analysis framework for the project, and establishing the analysis domain boundaries for intelligent auditing. This includes the following steps: The basic attributes of the project are extracted as parameters for the project profile, and the project application materials are deconstructed to form a multimodal representation of the material profile. Define the boundaries of the intelligent audit analysis domain, horizontally covering project progress status, material submission types, and on-site environmental data, and vertically setting analysis depth thresholds, including micro and macro levels; N monitoring indicators are set up at the control points, and M dynamic observation points are set up along the time dimension for each monitoring indicator; P control observation points are set up in the project reference system. By integrating the structured parameters of the project profile, the multimodal features of the material profile, the dynamic observation sequence of N monitoring indicators, and the benchmark data of the P control observation points, a digital profile of the project and a multidimensional feature analysis space are constructed.

[0014] In a preferred embodiment, the basic project attributes include the planned duration range, project type classification label, responsible department identifier, and initial risk level value; The multimodal representation includes semantic embedding vectors of text, numerical matrices of tables, and visual feature maps of images. The micro level includes the time deviation of individual milestone nodes, while the macro level includes the overall progress fluctuation pattern of departmental project groups.

[0015] This application also provides an artificial intelligence-based digital project management system, including: Basic Model Layer: Based on the needs of project lifecycle management, the project entities are digitally abstracted, a digital profile of the project and a multi-dimensional feature analysis framework are constructed, and the analysis domain boundaries of intelligent auditing are established; Computational Grid Layer: Based on the digital profile of the project, an intelligent analysis grid is automatically generated to adapt to multi-source heterogeneous data. The intelligent analysis grid includes a project progress grid and a material feature characterization grid. The project progress data is divided into grids according to the boundary of the analysis domain. The project grid is adaptively partitioned through an unsupervised clustering algorithm, and the grid density parameters are dynamically adjusted for each partition. Intelligent Decision Layer: Based on the digital profile of the project and the intelligent analysis grid, it comprehensively analyzes the risk stress, dynamically determines the critical threshold for risk warning, and outputs risk warning decision management conclusions that include the critical threshold.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This application digitizes and abstracts project entities (such as schedule nodes, material parameters, resource allocation, risk factors, and other core elements) to construct a comprehensive profile and multi-dimensional feature analysis framework covering all elements. It clarifies the analysis domain boundaries of intelligent auditing, solves the problems of scattered project status and difficulty in uniformly depicting features in the traditional model, and provides a structured and semantic data foundation for subsequent analysis, enabling managers to intuitively grasp the overall outline and key feature distribution of the project.

[0017] This application utilizes an intelligent analysis grid (including a project schedule grid and a material feature characterization grid) automatically generated from digital profiles. This grid is capable of adapting to the complex characteristics of multi-source heterogeneous data (such as time-series schedule records, unstructured material procurement documents, sensor monitoring data, etc.). It adaptively partitions the project schedule grid using an unsupervised clustering algorithm, accurately identifying key schedule clusters (such as critical path dense areas and lag risk areas) and material feature clusters (such as high-consumption material distribution areas and supply chain anomaly areas). Furthermore, it dynamically adjusts the grid density parameters based on the actual complexity of each partition. For high-risk or high-complexity partitions, the grid is refined to enhance the granularity of analysis, while the density is relaxed in low-risk areas to reduce computational redundancy. This approach improves computational efficiency while ensuring analytical accuracy.

[0018] This application combines the multi-dimensional features of a comprehensive digital profile of a project with the partitioning results of an intelligent analysis grid. It introduces a risk stress analysis model to dynamically calculate the risk impact weight and propagation path, breaking through the traditional static early warning mode with fixed thresholds. It can adjust the critical threshold of risk warning in real time according to the current status of the project (such as the degree of cumulative schedule deviation, the frequency of material supply fluctuations, and the intensity of resource conflicts), and output decision management conclusions that include dynamic thresholds. This enables early identification and precise intervention of potential risks, effectively avoiding the expansion of losses due to delayed early warnings. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0020] Figure 1 This is a framework diagram of the monitoring system of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0022] Example: This example provides a digital project management method based on artificial intelligence. The management method includes the following steps: S1. Project Profile Construction and Intelligent Analysis Space Initialization (Basic Model Layer) Based on the needs of project lifecycle management, we first construct a digital profile and multi-dimensional feature analysis framework for projects, establishing the analysis domain boundaries for intelligent auditing. We then perform digital abstraction of project entities: Extract basic project attributes (such as planned construction period range, project type classification label, responsible department identifier, and initial risk level value) as core parameters for project profiling. At the same time, perform feature deconstruction on project application materials (text / table / image / video, etc.) to form a multimodal representation of the material profiling (such as semantic embedding vector of text, numerical matrix of table, and visual feature map of image).

[0023] Define the analysis domain boundaries for intelligent auditing, horizontally covering project progress status (the entire timeline from project initiation to completion and acceptance), material application types (all scenarios such as delays / changes / completion), and on-site environmental data (such as timestamps and geographical location information of construction images); vertically set analysis depth thresholds, including micro-level (such as time deviation of a single milestone node) and macro-level (such as the overall progress fluctuation pattern of a department-level project group).

[0024] To accurately characterize the project, N monitoring indicators (corresponding to N types of core project indicators, such as schedule deviation rate and material completeness rate) are set at the key control points (i.e., the core stage of material submission deadline). For each monitoring indicator, M dynamic observation points are set along the time dimension (i.e., the project life cycle dimension). These observation points are M time series sampling points, with Q time units between each adjacent observation point. Q is dynamically adjusted according to the project type: Q=1 day for short-term (e.g., one week) projects, Q=7 days for medium-term (e.g., one month) projects, and Q=30 days for long-term (e.g., one year) projects. At the same time, P control observation points (P=M, corresponding to the benchmark values ​​of the same time dimension and indicator type, such as the historical average level of similar projects) are set in the project reference system (i.e., benchmark data with the same characteristic dimensions as the key control points but without direct interest).

[0025] By integrating the structured parameters of the project profile (such as construction period and type), the multimodal features of the material profile, the dynamic observation sequence of N monitoring indicators, and the benchmark data of P control observation points, a complete digital profile of the project and a multidimensional feature analysis space are constructed, establishing a precise analysis domain boundary for subsequent intelligent computing.

[0026] S2. Intelligent Grid Generation and Multimodal Data Adaptation (Computational Grid Layer) Based on the digital profile of the project constructed using S1, an intelligent analysis grid adapted to multi-source heterogeneous data is automatically generated. This grid contains two core units: a project schedule grid and a material characteristic characterization grid. First, the project schedule data is gridded according to the analysis domain boundary conditions (macro-micro analysis span, dynamic observation point time interval Q). The project timeline is discretized into several time step units (each unit corresponds to a progress status snapshot), and a three-dimensional mesh (time dimension × task dimension × resource dimension) is generated for each progress node. Key milestone nodes are used as high-precision mesh areas (the mesh density is dynamically adjusted according to the project risk level: the mesh density is increased by 30%-50% for high-risk projects).

[0027] To address the multimodal characteristics of material profiling, a feature representation grid is constructed for each material type (such as extension applications and completion reports). For text-based materials, a concept space grid is generated through semantic embedding vectors (each word / phrase corresponds to a grid node). For table-based materials, a structured grid is generated through numerical field mapping (each cell corresponds to a numerical feature node). For image / video-based materials, a visual grid is generated through computer vision feature extraction (each pixel / frame corresponds to a feature node).

[0028] The dynamic observation sequences of N monitoring indicators (each indicator generates a time series grid) are further integrated with the baseline grid of P control observation points (historical characteristic grids of similar projects) to form a composite analysis grid system. This grid system supports cross-modal data correlation (such as correlation analysis between milestone delay nodes in the schedule grid and extension application submission time nodes in the materials grid) and multi-scale feature extraction (from the micro-state of a single task to the overall trend of a group of projects).

[0029] To adapt to the different characteristics of different project types, the project grid is adaptively partitioned using unsupervised clustering algorithms (such as grouping based on project duration). The grid density parameters are dynamically adjusted for each partition (e.g., the time step of the schedule grid for short-term construction projects is set to 1 day, and the time step of the grid for long-term information projects is set to 7 days). Based on the distribution of typical issues in the historical audit case library (e.g., schedule deviation issues are concentrated within 3 days before and after key milestones, and material shortage issues frequently occur within 24 hours before the submission deadline), the monitoring weight of grid nodes is optimized (e.g., the grid weight of key nodes is increased by 20%-40%).

[0030] S3. Dynamic Threshold Risk Perception and Intelligent Decision-Making (Intelligent Decision-Making Layer) Based on the project's digital profile (S1) and intelligent analysis grid (S2), multimodal data fusion and dynamic threshold adjustment are used to analyze the interaction risk patterns between project progress and materials, and to determine the critical threshold for digital management risk warning. First, an equivalent scenario of key node advancement is simulated, treating the project progress update process as a dynamic influencing factor acting on core control points. By calculating the response strength (e.g., timeliness of material submission) of material application data (e.g., extension applications) when project status changes (e.g., updates to task completion rates), a risk amplitude distribution (i.e., the risk sensitivity of each monitoring point) is generated.

[0031] Based on the analysis grid (schedule grid and material characteristic characterization grid) and the risk interaction model (the nonlinear coupling effect between project schedule and materials, such as the synergistic risk of schedule delays and material shortages), the risk amplitude (quantifying risk intensity) of each monitoring point is calculated, and a risk distribution cloud map is synthesized (visualizing the risk distribution pattern of each node). Subsequently, the flow field structure of the risk distribution cloud map is analyzed. Identify high-risk clusters (such as nodes where schedule delays and material shortages co-occur), abrupt risk gradient zones (such as transition nodes from normal schedules to high-risk delays), and risk eddies (such as schedule stalls in cross-departmental collaborations). Based on the flow field structure analysis results, further conduct risk force analysis, mapping risk amplitudes to specific risk factors in project management (such as the risk of compliance review failure due to material shortages), and calculate the intensity (quantified based on the probability of issue association and degree of impact in historical audit cases) and direction of action (such as positively promoting project progress or negatively hindering compliance).

[0032] By comprehensively analyzing the risk forces, the critical threshold for risk warning is dynamically determined: For the progress management dimension, if the risk amplitude of a certain milestone node exceeds the dynamic critical threshold (the threshold is automatically adjusted according to the project type: 0.85 for construction projects and 0.75 for information technology projects), an early warning will be triggered (indicated by the need to submit complete materials X days before the milestone, where X is automatically calculated based on the risk gradient). For the material management dimension, if the risk magnitude of a certain element of the material feature characterization grid (such as signature integrity) exceeds the dynamic critical threshold (the threshold is dynamically adjusted based on the missing rate of the element in historical rejection cases), it is marked as a high-risk missing item (the backoff distance is expressed as the requirement to complete it Y hours before the submission deadline, where Y is quantified according to the urgency).

[0033] The final output includes risk warning decision management conclusions with dynamic thresholds (e.g., a completion report with the owner's signature must be submitted 3 days in advance for key milestones; if the progress is delayed by more than 10%, an explanation of the reasons and a remedial plan must be submitted simultaneously).

[0034] like Figure 1 As shown: This embodiment also provides an artificial intelligence-based digital project management system, including: Basic Model Layer: Based on the needs of project lifecycle management, the project entity is digitally abstracted, a digital profile of the project and a multi-dimensional feature analysis framework are constructed, the analysis domain boundary of intelligent review is established, the analysis domain boundary is sent to the computing grid layer, the digital profile of the project is sent to the computing grid layer and the intelligent decision layer; Computational Grid Layer: Based on the digital profile of the project, an intelligent analysis grid is automatically generated to adapt to multi-source heterogeneous data. The intelligent analysis grid includes a project progress grid and a material feature characterization grid. The project progress data is divided into grids according to the boundary of the analysis domain. The project grid is adaptively partitioned through an unsupervised clustering algorithm. The grid density parameters are dynamically adjusted for each partition. The intelligent analysis grid is sent to the intelligent decision layer. Intelligent Decision Layer: Based on the digital profile of the project and the intelligent analysis grid, it comprehensively analyzes the risk stress, dynamically determines the critical threshold for risk warning, and outputs risk warning decision management conclusions that include the critical threshold.

[0035] The following is a more detailed and publicly disclosed description of each implementation step of this application: Furthermore, basic attributes are extracted from project application information as core parameters for project profiling. These parameters are key dimensions for defining project identity and initial risk characteristics. Specifically, they include: The planned project duration range (e.g., 3 to 12 months, reflecting the expected project timeframe), project type classification tags (e.g., construction engineering, IT system development, equipment procurement, etc., used to differentiate project management logic across different areas), responsible department identifiers (e.g., engineering department, R&D center, etc., clearly defining the management entity), and initial risk level (an initial risk score based on historical data from similar projects, such as low / medium / high risk levels). These structured parameters form the basic framework of the project profile, providing metadata support for subsequent analysis.

[0036] Furthermore, the project application materials are simultaneously subjected to multimodal feature deconstruction, and differentiated feature representations are extracted for data in different formats: Text-based materials are generated into semantic embedding vectors using natural language processing techniques (e.g., 768-dimensional text semantic features extracted based on the BERT model to capture deep semantic relationships of key information such as project goals and technical requirements); tabular materials are parsed into numerical matrices (preserving original numerical values ​​and row and column relationships, such as structured data like material procurement quantities and budget allocation ratios, used for quantitative analysis of resource demand and cost distribution); image-based materials are extracted into visual feature maps using computer vision techniques (e.g., key area feature vectors obtained through the ResNet model, focusing on visual information such as spatial layout and equipment location in drawings); and video-based materials are generated into dynamic feature sequences through temporal analysis (e.g., temporal frame features of key operations, used to capture the continuous characteristics of the construction process).

[0037] Furthermore, based on the needs of project lifecycle management, it is clear that the analytical domain boundary of intelligent auditing needs to cover both the horizontal scenario dimension and the vertical depth dimension: The horizontal coverage includes project progress status (the entire timeline from the planning and initiation stage to the completion and acceptance stage, covering key stages such as project initiation, design, construction, and acceptance), material submission types (such as extension applications, change applications, completion reports, etc., with differentiated review rules set for different submission types), and on-site environmental data (such as timestamps and geographical location information of construction images, used to associate specific construction behaviors with spatial locations, and assist in judging the compliance and timeliness of on-site operations).

[0038] The vertical depth threshold sets the analytical granularity at both the micro and macro levels. The micro level focuses on the fine characteristics of individual milestone nodes (such as the deviation rate between the planned and actual completion time of a critical path node, or the difference between the actual and planned quantities of a material batch). The macro level focuses on the overall pattern of a group of projects (such as the consistency of progress fluctuations of multiple parallel projects at the department level, or the distribution pattern of cost overrun rates of similar projects).

[0039] By defining both horizontal and vertical boundaries, intelligent auditing can capture both local anomalies and identify overall trend risks.

[0040] Furthermore, to accurately characterize the project, key control points (i.e. links that have a decisive impact on the success of the project, such as material submission deadlines, milestone acceptance nodes, and other core links) are selected in the project life cycle, and N monitoring indicators are set up (each indicator corresponds to a type of core project indicator, such as schedule deviation rate, material integrity rate, cost overrun rate, etc.).

[0041] For each monitoring indicator, M dynamic observation points (i.e., time-series sampling points) are set along the project lifecycle dimension to capture the trend of indicator changes over time. The time interval Q between adjacent observation points is dynamically adjusted according to the project type. For short-term projects (such as urgent tasks with a duration of ≤1 week), Q=1 day (high-frequency monitoring, tracking indicator changes daily). For medium-term projects (such as regular projects with a construction period of 1 to 3 months), Q = 7 days (sampling every week to balance monitoring accuracy and calculation cost). For long-term projects (such as large-scale projects with a construction period of ≥1 year), Q=30 days (monthly sampling, focusing on phased progress).

[0042] Example: For an IT system development project with a duration of 6 months (mid-term project), if N=3 monitoring indicators are set (schedule deviation rate, code submission completeness rate, test pass rate), then M=4 (6 months ≈ 24 weeks, sampled once a week). Each indicator will generate 4 time series observation points (week 1, week 2, ..., week 4) to analyze the time series fluctuation characteristics of the indicator.

[0043] Simultaneously, P control observation points (P=M, completely consistent with the monitoring indicator types and time dimensions of the key control points) are set up in the project reference system. These control points are derived from benchmark data with no direct interest (such as the average level of similar historical projects, industry benchmark values, or statistical values ​​of project groups in similar departments). For example, for the schedule deviation rate indicator of the current project, the average schedule deviation rate of the first week in similar IT projects over the past 3 years can be selected as the benchmark value for the control observation point in week 1.

[0044] Furthermore, by integrating multi-source data, a complete digital profile of the project and a multi-dimensional feature analysis space are constructed: The structured parameter layer integrates basic project attributes (planned duration range, project type tags, etc.) to provide project identification and initial management parameters; the multimodal material layer embeds multimodal features of application materials (text semantic vectors, table numerical matrices, etc.) to reflect detailed information and technical details of project requirements; the dynamic monitoring layer contains M time-series observation points for N monitoring indicators (such as the progress deviation rate in week 1 and week 2) to characterize the temporal evolution of key project indicators; the benchmark reference layer introduces benchmark data for P control observation points (such as the historical progress deviation rate of similar projects) for horizontal comparative analysis of the current project's deviation degree.

[0045] Furthermore, the core objective of the project schedule grid is to transform the progress status on the project timeline into computable three-dimensional spatial units. Its generation process consists of the following key steps: Based on the planned duration range (e.g., 3 to 12 months) and dynamic observation point time intervals Q (Q=1 day for short-term projects, Q=7 days for medium-term projects, and Q=30 days for long-term projects) in the project's digital profile, the project's entire lifecycle timeline is discretized into several time step units (e.g., a 6-month medium-term project will generate approximately 24 time units if divided at Q=7 days). Each time step unit corresponds to a progress status snapshot, recording the planned and actual progress of all tasks at that moment (e.g., task completion percentage, resource input).

[0046] Each time step unit is further unfolded into a three-dimensional mesh structure, the dimensions of which are defined as: The time dimension corresponds to the discretized time step units (e.g., week 1, week 2, etc.), used to anchor the time point of the progress status; the task dimension covers all sub-tasks in the Work Breakdown Structure (WBS) (e.g., the design phase includes sub-tasks such as requirements analysis and scheme design), with each task as an independent dimension node; the resource dimension associates the resource types required for task execution (e.g., human resources, equipment resources, material resources), with each resource type as a dimension node. Each cell of this three-dimensional grid (time i, task j, resource k) stores triplet feature values: the amount of resource k invested in task j at time i (e.g., 3 engineers, 2 excavators), the actual progress (e.g., 60%), and the planned progress (e.g., 50%), thereby characterizing the progress status.

[0047] For key milestones in the project (such as the completion of the main structure's topping-out or the completion of system integration testing), their corresponding grid cells are marked as high-precision grid areas. The dynamic adjustment logic for the grid density of these nodes is as follows: The grid density of ordinary nodes is determined by the time step unit Q (e.g., when Q=7 days, each milestone node is associated with 1 time unit grid); based on the initial risk level value in the project's digital profile (e.g., high risk, medium risk, low risk), the high-precision grid density of high-risk projects (risk level ≥ 80 points) is increased by 30%-50% (e.g., nodes with Q=7 days are adjusted to Q=5 days or Q=3 days, increasing the number of time step units), medium-risk projects are increased by 15%-25%, and low-risk projects maintain the basic density unchanged.

[0048] Example: A construction project was assessed as high-risk (risk level 85). The original milestone node for the completion of the main structure was divided into Q=7 days (1 grid unit). After adjustment, it was densified to Q=3 days (about 2 additional grid units) to more intensively monitor the resource input and schedule deviation of this node.

[0049] Furthermore, the material feature characterization grid constructs a differentiated feature space grid for each material type (such as extension applications and completion reports) based on the multimodal characteristics (text / tables / images / videos) of the project application materials, as detailed below: For text-based materials (such as project feasibility reports and requirements specifications), natural language processing techniques are used to extract semantic features. First, a pre-trained language model (such as BERT) is used to transform the text content into semantic embedding vectors (e.g., each sentence or paragraph is mapped to a 768-dimensional vector). Then, based on semantic similarity clustering, the vector space is divided into several concept clusters (e.g., topic clusters such as technical requirements, schedules, and risk warnings). Each word / phrase (or the center of the clustered topic cluster) serves as a grid node, and the connection weights between nodes are determined by semantic relevance (e.g., cosine similarity), forming a concept space grid. This grid can intuitively reflect the distribution and association of key information in the material (e.g., cost control-related words are concentrated in a certain area of ​​the grid).

[0050] For tabular documents (such as budget tables and resource allocation tables), parse their row and column structure and extract numerical fields: Each row (record) and column (field) of the table is mapped to row and column nodes of a grid. Cells (row i, column j) store raw numerical characteristics (e.g., the budget value for material A in row 3 is 500,000 yuan, or the total allocation of human resources in column 2 is 10 person-months). Through numerical normalization (e.g., scaling to the 0-1 range using Min-Max), the comparability of fields with different dimensions is ensured, forming a structured grid. This grid supports rapid location and analysis of quantitative characteristics such as resource allocation and cost distribution (e.g., directly viewing the numerical distribution of the equipment procurement column).

[0051] For image-based materials (such as construction drawings and design renderings) and video-based materials (such as construction process recordings and equipment debugging videos), computer vision technology is used to extract visual features: Image materials extract key region features (such as equipment layout areas and building structure outlines in drawings) from images using convolutional neural networks (such as ResNet). Each significant pixel region (such as high-contrast regions after edge detection) or predefined interest point (such as annotation boxes in design drawings) is used as a grid node, and the node features are the visual description vectors of the region (such as color histograms and texture features). The video material is processed frame by frame. Visual features (such as time-series frames of key operations) are extracted from each frame. Each video frame (or key frame) is used as a grid node, and the node features are the visual description vectors of that frame (such as the feature vectors of construction workers' actions). Finally, a visual grid is formed, and the spatial distribution of its nodes reflects the key areas of visual information (for example, areas that are frequently modified in drawings show high-density node clusters in the grid).

[0052] Furthermore, the project schedule grid tracks the dynamic process of project execution through a three-dimensional structure of time, task, and resources, while the material feature characterization grid analyzes the inherent semantics and key information of the application materials through a multimodal feature space. Both types of grids strictly adhere to the boundary conditions of the analysis domain (e.g., the macro-micro analysis span determines the time resolution and feature granularity of the grid, and the time interval Q of dynamic observation points directly affects the partitioning density of time step units), and achieve cross-grid correlation queries through a data indexing mechanism (e.g., when locating the task progress of a certain time step unit, the key information in the material feature grid submitted during that period is simultaneously correlated).

[0053] Furthermore, for each monitoring indicator (such as schedule deviation rate, material integrity rate, and cost overrun rate), a time-series grid is generated based on M dynamic observation points (time-series sampling points) set along the project lifecycle. Specifically: The M observation points for each monitoring indicator are arranged chronologically to form a one-dimensional time series data (e.g., the sequence of values ​​for the schedule deviation rate in week 1, week 2, ..., week M). This sequence is mapped to the time dimension of the project schedule grid using a time step alignment algorithm: based on the discretized time step unit of the project timeline (e.g., Q = 7 days per unit), the timestamp of the observation point (e.g., Monday of week 2) is matched to the nearest time step unit (e.g., the unit corresponding to week 2), thus transforming the one-dimensional time series into a two-dimensional grid structure (time dimension × indicator dimension) that shares the time dimension with the schedule grid. Each grid node stores the specific value of a monitoring indicator within that time step unit (e.g., the schedule deviation rate for week 2 is +12%).

[0054] Furthermore, the independent time series grids of N monitoring indicators are spliced ​​together according to the indicator dimension to form an N-dimensional time series grid set (time dimension × task dimension × resource dimension × indicator 1... indicator N). For example, if N=3 (schedule deviation rate, material integrity rate, test pass rate), then each time step unit will contain numerical nodes of 3 indicators, which are used to synchronously analyze the coordinated changes of multiple indicators (such as whether schedule delays are accompanied by material shortages or test failures).

[0055] Example: For the schedule deviation rate indicator, the value of its third observation point (week 3) is +8%. It is matched to the third week cell of the project schedule grid through the time step alignment algorithm, and finally the schedule deviation rate = +8% is stored in the indicator dimension node of the cell. Similarly, the value of the material integrity rate indicator in week 3 is 92% (8% of the materials are missing and have not been submitted), and it is synchronously stored in the corresponding node of the same time cell.

[0056] To provide a basis for horizontal comparison, a baseline grid of P control observation points (P=M, completely consistent with the number and time dimension of dynamic observation points) is generated based on the project reference system (similar benchmark data with no direct interest). Select similar projects (such as similar construction projects or IT system development projects of the same scale) from the historical project database that share the same characteristics as the current project. Extract the historical average (or quantiles, such as P50 / P75) and standard deviation of their key control points at the same time step (e.g., week 1, week 2, ..., week M) to form a historical feature grid for the similar project group. For example, for the schedule deviation rate indicator, select the average schedule deviation rate of +5% (standard deviation ±3%) for the second week of all similar construction projects over the past 3 years as the benchmark reference value for the second week of the current project.

[0057] By aligning historical feature grids of similar project groups with the index dimension according to time step units, a baseline grid set (time dimension × index dimension) consistent with the grid structure of the dynamic observation sequence is generated. Each baseline grid node stores historical statistical values ​​(such as mean, quantiles) or extreme value thresholds (e.g., values ​​exceeding P90 are considered outliers) for subsequent deviation calculations. Example: If the current project is a medium-sized IT system development (feature dimension matching), its code commit integrity rate in week 4 is obtained by aggregating the data of week 4 of similar historical projects, and the baseline mean is 85% (standard deviation ±5%). Then the code commit integrity rate of the node storage in week 4 of the baseline grid is 85% ± 5%.

[0058] Furthermore, the time-series grid set of N monitoring indicators and the baseline grid set of P control observation points are topologically integrated with the original project schedule grid and material characteristic characterization grid to form a composite analysis grid system. The integration logic of this system is as follows: Each grid node within a time step cell is simultaneously associated with four types of data: Task-resource status in the schedule grid (e.g., design task completed 60% in week 3, 2 engineers deployed); multimodal features in the materials grid (e.g., semantic embedding vectors of extension application materials submitted in week 3, reflecting keywords of the application reason); time series grid of dynamic observation sequences (e.g., schedule deviation rate +8% and material integrity rate 92% in week 3); and comparative data from the baseline grid (e.g., the baseline value of schedule deviation rate in week 3 for similar projects is +5% ± 3%).

[0059] Furthermore, data association between different grids is achieved through multimodal index labels: For example, a unique identifier is added to the milestone delay node in the schedule grid (such as the main structure acceptance delayed by 2 weeks). At the same time, the delay application materials submitted within the same time interval are retrieved in the material grid (by matching timestamps) and their semantic features are extracted (such as the application reason containing the keyword of material supply delay), thereby establishing a causal link between schedule anomalies and material behavior.

[0060] Further, the multi-scale feature extraction hierarchy: The micro-scale focuses on the local state of a single task or type of material (e.g., insufficient resource allocation for a subtask in week 2 leading to a 15% schedule deviation rate, or excessively frequent modifications to key areas in the visual grid of a design drawing); the macro-scale captures the overall trend of a group of projects (e.g., schedule fluctuations in multiple parallel projects at the departmental level exhibiting a synchronous delay pattern, or a right-skewed anomaly in the distribution of cost overruns for similar projects). Example: In macro-scale analysis, by aggregating the schedule deviation rate benchmark grid of all similar projects (P control observation points), the industry average schedule fluctuation curve is calculated (e.g., the average deviation rate of weeks 1-4 is ≤+3%, and the deviation rate rises to +8% after week 5). The current project's schedule grid is compared with this curve. If the deviation rate of week 3 has reached +12% (significantly higher than the industry P75 quantile), a macro-risk warning is triggered.

[0061] Furthermore, key clustering features are extracted from the digital profile of the project, including but not limited to: The project duration range (e.g., 3 to 12 months), task complexity indicators (e.g., the number of sub-tasks in the WBS decomposition, the number of cross-departmental collaboration nodes), resource input intensity (e.g., average daily manpower, equipment utilization rate), and historical risk characteristics (e.g., the mean schedule deviation rate and the quantile of the material shortage rate for similar projects) are all considered. After standardization (e.g., Min-Max normalization to the range of [0,1]), these characteristics are used to construct a multi-dimensional feature vector for the project (e.g., [normalized value of duration, normalized value of task complexity, normalized value of resource intensity, normalized value of historical risk]).

[0062] Density-based clustering algorithms (such as DBSCAN) are used to automatically group the item set based on the similarity of the item feature vectors (such as Euclidean distance or cosine similarity). For example: The DBSCAN algorithm clusters densely connected projects in the feature space into groups by setting a neighborhood radius (eps) and a minimum number of samples (min_samples). For example, projects with a duration ≤ 3 months and low task complexity are grouped into a short-term, simple project cluster, while projects with a duration ≥ 9 months and intensive cross-departmental collaboration are grouped into a long-term, complex project cluster. The clustering results output several project clusters (denoted as C1, C2, ..., C...). k k represents the number of clusters, and each cluster represents a type of project with similar management needs.

[0063] Further, the mapping logic for partitioning rules: Projects within the same cluster share the same partition parameter configurations (such as grid density and time step units). For example, if cluster analysis finds that a company's project set can be divided into three clusters: Cluster A (short-term construction projects, duration ≤ 3 months), Cluster B (medium-term IT projects, duration 3-9 months), and Cluster C (long-term system integration projects, duration ≥ 9 months), then subsequent grid parameter adjustments will be performed separately for these three partitions.

[0064] Based on the clustering partitioning results, the time step unit of the grid (i.e., the time resolution of the schedule grid) is dynamically adjusted for each project partition. The data collection granularity is optimized by mapping the duration of the project to risk sensitivity, thus balancing monitoring accuracy and computational efficiency.

[0065] Furthermore, define the relationship between the basic time step unit and the project duration: For short-term project clusters (e.g., duration ≤ 3 months), the time step unit is set to 1 day (high-frequency monitoring, daily tracking of task progress and resource input); for medium-term project clusters (e.g., duration 3-9 months), the time step unit is set to 7 days (weekly sampling, covering regular progress update cycles); and for long-term project clusters (e.g., duration ≥ 9 months), the time step unit is set to 30 days (monthly sampling, focusing on the progress of phased milestones).

[0066] Furthermore, the risk level will be dynamically adjusted: If the initial risk level in the project's digital profile exceeds the threshold (e.g., for high-risk projects, the risk score is ≥80), then the basic time step unit will be compressed and adjusted. For short-term high-risk projects, the time step unit is shortened from 1 day to 0.5 days (monitoring every other day), or it is kept at 1 day but the grid density of key nodes is increased (such as adding extra time units around key milestones); for medium-term high-risk projects, the time step unit is shortened from 7 days to 3-5 days (sampling once every 3-5 days); for long-term high-risk projects, the time step unit is shortened from 30 days to 15-20 days (monitoring once every half month).

[0067] Example: A municipal road construction project (2-month construction period, belonging to cluster A) is assessed as high risk (risk score 85 points). Its progress grid time step unit is further compressed from the basic value of 1 day to 0.5 days (monitored twice daily), focusing on tracking the daily progress of key processes such as earthwork excavation and concrete pouring; while an enterprise resource planning (ERP) system development project (8-month construction period, belonging to cluster B) is of medium risk (risk score 60 points), so a basic time step unit of 7 days is adopted (sampling weekly).

[0068] Based on the distribution patterns of typical issues in the historical audit case library (such as progress deviations concentrated around key milestones and missing materials frequently occurring before deadlines), the monitoring weight of grid nodes is dynamically adjusted through an issue hotspot localization algorithm, prioritizing high-risk periods and key control points.

[0069] Furthermore, the spatiotemporal distribution characteristics of two typical issues were extracted from the historical review case database: Statistically analyze the time stamps of nodes where the progress deviation rate exceeds the threshold (e.g., ±10%), and analyze their time offset relative to key milestones (e.g., 80% of the progress deviation occurs within 3 days before the milestone starts and 2 days after its completion); statistically analyze cases of overdue material submissions and record their time distribution before the submission deadline (e.g., 75% of the overdue events occur within 24 hours before the deadline).

[0070] Furthermore, based on the above statistical results, the determination rule for high-weight grid nodes is defined as follows: Key milestones are nodes marked as critical paths in the project breakdown structure (WBS) (such as the main structure being capped or system commissioning), or nodes where the schedule deviation tolerance is below a threshold (such as allowable deviation ≤ ±5%). High-risk time windows are time intervals with concentrated schedule deviation issues (such as 3 days before and after a milestone) and periods with high frequency of missing materials (such as 24 hours before the submission deadline).

[0071] Furthermore, for the identified high-weight grid nodes, their monitoring priority is increased through a weight adjustment function: Base weight (all grid nodes have a default weight of 1.0); the weight of critical milestone nodes is increased by 20%-40% (e.g., from 1.0 to 1.2-1.4) to enhance tracking of their progress status and resource allocation; grid nodes within high-risk time windows (e.g., time units 3 days before and after a milestone) have their weights increased by an additional 10%-20% (e.g., a total weight of 1.3-1.6 after aggregation); low-weight nodes (e.g., regular task nodes on non-critical paths) maintain their base weight or are appropriately reduced (e.g., reduced to 0.8-0.9) to reduce unnecessary computational overhead. Example: The installation of the operating room purification system in a hospital expansion project is a key milestone node (allowable deviation ≤ ±3%). The weight of its corresponding progress grid node is increased from 1.0 to 1.4. At the same time, the weight of the time unit to which the deadline for submitting materials (such as the purification equipment acceptance report) for this node belongs is increased to 1.3, which is used to focus on monitoring the timeliness of material submission.

[0072] Furthermore, for each key node in the project schedule grid (such as design review approval, main structure completion), the status update process is simulated (e.g., task completion rate increases from 70% to 90%), and the changes in the application data of related materials in the material characteristic representation grid (such as the construction drawings and equipment procurement list corresponding to the node) are observed simultaneously (e.g., the submission time of extension application, the quantity of missing materials). Through the status-response correlation analysis algorithm, the correlation strength between the change in task status (Δ task progress = current completion rate - previous cycle completion rate) and the material response (Δ material response = material submission delay days or material shortage rate change) is calculated, generating the initial risk amplitude for each monitoring point (for example, if a node's schedule is delayed by 10% and accompanied by a 5-day delay in the submission of a material extension application, then the initial risk amplitude for that node is highly sensitive).

[0073] Furthermore, the quantification of risk magnitude: Based on multidimensional data from an intelligent analysis grid (task-resource status in the schedule grid, multimodal features in the material grid, and index values ​​in the dynamic observation sequence), the risk amplitude (quantifying risk intensity) of each monitoring point is calculated using a nonlinear coupling effect analytical algorithm. Specifically: Extract the time deviation rate of a node in the schedule grid (e.g., actual progress is 15% behind schedule) and the integrity rate of the corresponding material in the material grid (e.g., missing key drawings result in a material integrity rate of 70%). Combine the schedule deviation rate index (e.g., +15%) and the material integrity rate index (e.g., the decrease from 92% to 70%) of that node in the dynamic observation sequence. Generate a comprehensive risk amplitude through multi-dimensional feature weighting (e.g., schedule deviation weight 0.6, material integrity rate weight 0.4). (For example, a node with a 15% delay in schedule and a 22% decrease in material integrity rate is marked as high risk.)

[0074] Furthermore, to intuitively present the risk distribution characteristics of each node throughout the project's entire lifecycle, the risk amplitudes of each monitoring point are synthesized into a risk distribution cloud map (a three-dimensional visualization matrix containing time, task, and risk intensity dimensions), and the topology of high-risk areas is further identified through flow field structure analysis algorithms.

[0075] Based on the time step units (e.g., weekly or monthly) and task breakdown structure (WBS) of the intelligent analysis grid, the risk amplitude of the monitoring point corresponding to each time-task unit is mapped to the cloud map coordinate system: the horizontal axis is the time dimension (e.g., week 1 to week 24 of the project), the vertical axis is the task dimension (e.g., design phase, construction phase, acceptance phase), and the color depth (or height) represents the risk amplitude intensity (e.g., red represents high risk, green represents low risk). For example: If a project's main structure construction node in week 5 is simultaneously delayed by 12% and has a steel reinforcement material shortage rate of 30%, then that node will be displayed as a dark red high-density area in the cloud map.

[0076] Furthermore, by calculating the risk magnitude gradient (the derivative in the spatial / temporal direction) of each node in the risk distribution cloud map, the following three typical risk topologies are identified: High-risk clusters are continuous areas where the risk amplitude is significantly higher than that of neighboring nodes (such as the co-occurrence of delays and missing materials at multiple key milestone nodes, forming a cluster of risk hotspots). For example, in the system commissioning and final acceptance of a project, both of the two consecutive milestone nodes experienced delays of more than 10% and material submissions were overdue by more than 5 days. In the cloud map, this area appears as a continuous, high-density red distribution.

[0077] A risk gradient abrupt change zone is a transitional node where the risk magnitude changes drastically within a short period or a small range (such as a node where project progress suddenly jumps from normal (deviation rate ≤ 5%) to high-risk delay (deviation rate ≥ 15%), or a critical point where material integrity drops sharply from 90% to 60%). For example, if a construction node is delayed by 5 days due to sudden weather, and the material supply is completely interrupted on the second day (integrity rate drops from 80% to 20%), this node constitutes a gradient abrupt change zone.

[0078] A risk vortex zone is a node where the risk amplitude accumulates cyclically within a specific area and is difficult to dissipate (e.g., in cross-departmental collaboration, communication delays lead to progress stagnation, which in turn causes repeated material submission errors, forming a negative feedback loop between progress and materials). For example, a delay in the drawing confirmation process between the design and construction departments causes construction progress to stagnate (the progress deviation rate increases by 3% per week), while errors in the drawing version lead to repeated revisions of the material procurement list (the material integrity rate remains below 70%). This node appears as a vortex-shaped red distribution in the cloud map.

[0079] Furthermore, the risk magnitude is mapped to specific risk factors in project management (such as failure to pass compliance review, escalation of resource conflicts), and the critical threshold for risk warning is dynamically determined by quantifying the intensity and direction of each risk factor. Calculation of the intensity and direction of risk factors: Extract the correlation probability (e.g., the historical probability of compliance review failure due to missing materials is 65%) and impact degree (e.g., the average number of days the issue causes project delays is 5 days) from the historical audit case database. Define the strength of effect (strength value = correlation probability × impact degree, normalized to the range of 0-1) and the direction of effect (positive: promotes project progress, such as submitting materials in advance to speed up approval; negative: hinders project goals, such as delayed submission leading to compliance risks) for each risk factor (e.g., missing materials, schedule delays, cross-departmental collaboration failures). For example: The risk factor that delays the schedule by more than 10% has an impact strength of 0.8 (80% correlation probability, 10-day delay in impact) and is negative; the risk factor that ensures the material integrity rate is ≥95% has an impact strength of 0.3 (40% correlation probability, 2-day acceleration in impact) and is positive.

[0080] Furthermore, considering the intensity and direction of all risk factors, a comprehensive risk value for each monitoring point is calculated using a risk contribution weighting algorithm (risk value of a node = schedule delay intensity × 0.6 + material shortage intensity × 0.4), combined with the topological characteristics of the risk distribution cloud map (e.g., the risk value of high-risk clusters needs to be weighted an additional 10%-20%). Finally, based on the project type (short-term / long-term), risk level (high / medium / low), and management objectives (e.g., maximum tolerable delay days), a critical threshold for risk warning is dynamically set. For nodes in high-risk clusters or gradient mutation zones, the critical threshold is set to the P80-P90 quantile of the comprehensive risk value (an early warning is triggered when the comprehensive risk value of a node exceeds the 85th quantile of historical data for similar nodes); for nodes in risk eddies, the critical threshold is further reduced by 10%-15% (risks accumulate faster due to cross-departmental collaboration issues, requiring early intervention); for monitoring at the macro-level project cluster level (such as the overall progress fluctuation of multiple parallel projects at the departmental level), the critical threshold is dynamically adjusted based on the mean ± standard deviation of the historical risk distribution of similar projects (a group risk warning is triggered when the average progress deviation rate of the project cluster exceeds the P75 quantile + 1 standard deviation), example: In the 12th week, a smart city construction project identified two core risk factors at its data center deployment node through risk stress analysis: a 12% delay in schedule (intensity 0.9, negative) and a 40% missing material (server equipment list) rate (intensity 0.7, negative). The combined risk value reached 0.82 (exceeding the historical P85 quantile of 0.78 for similar nodes). At the same time, this node was located in the risk gradient abrupt change zone (the schedule jumped from normal to high-risk delay). Therefore, the system dynamically set the critical threshold to 0.80, triggering a level-one risk warning (requiring the submission of a schedule correction plan and a material supplement plan within 48 hours).

[0081] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0082] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A digital project management method based on artificial intelligence, characterized in that: The management method includes the following steps: S1. Based on the needs of project lifecycle management, the project entity is digitally abstracted, a digital profile of the project and a multi-dimensional feature analysis framework are constructed, and the analysis domain boundary of intelligent audit is established; S2. Based on the digital profile of the project, an intelligent analysis grid is automatically generated to adapt to multi-source heterogeneous data. The intelligent analysis grid includes a project progress grid and a material feature characterization grid. The project progress data is divided into grids according to the boundary of the analysis domain. The project grid is adaptively partitioned by an unsupervised clustering algorithm, and the grid density parameters are dynamically adjusted for each partition. S3. Based on the project's digital profile and intelligent analysis grid, comprehensively analyze the risk stress, dynamically determine the critical threshold for risk warning, and output risk warning decision management conclusions that include the critical threshold.

2. The digital project management method based on artificial intelligence according to claim 1, characterized in that: In step S3, based on the project's digital profile and intelligent analysis grid, and through comprehensive risk stress analysis, the critical threshold for risk warning is dynamically determined, including the following steps: The equivalent scenario of simulating node advancement is used to compare the project progress update process to dynamic influencing factors acting on control points. By calculating the response intensity of material declaration data when the project status changes, the risk amplitude distribution is generated. Based on the analysis grid and risk interaction model, the risk amplitude of each monitoring point is calculated, and a risk distribution cloud map is synthesized. The flow field structure of the risk distribution cloud map is then analyzed. Identify risk clustering areas, risk gradient abrupt change zones, and risk vortex regions. Based on the flow field structure analysis results, map the risk amplitude to project management risk factors and calculate the intensity and direction of each risk factor. By comprehensively analyzing the risk forces, the critical threshold for risk warning is dynamically determined.

3. The digital project management method based on artificial intelligence according to claim 2, characterized in that: The output includes risk warning decision management conclusions containing critical thresholds, comprising the following steps: In terms of progress management, if the risk magnitude of a certain milestone node exceeds the dynamic critical threshold, an early warning will be triggered.

4. The digital project management method based on artificial intelligence according to claim 2, characterized in that: The output includes risk warning decision management conclusions containing critical thresholds, comprising the following steps: For the materials management dimension, if the risk amplitude of a certain element in the material characteristic characterization grid exceeds the dynamic critical threshold, it is marked as a risk missing item.

5. The digital project management method based on artificial intelligence according to claim 1, characterized in that: In step S2, based on the project's digital profile, an intelligent analysis grid adapted to multi-source heterogeneous data is automatically generated, including the following steps: The project schedule data is divided into grids based on the boundary conditions of the analysis domain. The project timeline is discretized into several time step units, and a three-dimensional grid is generated for each schedule node, represented as: time dimension × task dimension × resource dimension.

6. The digital project management method based on artificial intelligence according to claim 5, characterized in that: In step S2, based on the project's digital profile, an intelligent analysis grid adapted to multi-source heterogeneous data is automatically generated, including the following steps: To address the multimodal characteristics of material profiling, a feature representation grid is constructed for each material type. For textual materials, a concept space grid is generated through semantic embedding vectors; for tabular materials, a structured grid is generated through numerical field mapping; and for image / video materials, a visual grid is generated through computer vision feature extraction.

7. The digital project management method based on artificial intelligence according to claim 6, characterized in that: In step S2, the project progress data is divided into grids according to the analysis domain boundaries. An unsupervised clustering algorithm is used to adaptively partition the project grid, and the grid density parameters are dynamically adjusted for each partition. This includes the following steps: A composite analysis grid system is formed by integrating the dynamic observation sequences of N monitoring indicators with the baseline grid of P control observation points; The project grid is adaptively partitioned using an unsupervised clustering algorithm. The grid density parameter is dynamically adjusted for each partition, and the monitoring weight of the grid nodes is optimized based on the issue distribution in the historical audit case library.

8. The digital project management method based on artificial intelligence according to claim 1, characterized in that: In step S1, based on the requirements of project lifecycle management, the project entity is digitally abstracted, a digital profile of the project and a multi-dimensional feature analysis framework are constructed, and the analysis domain boundary of intelligent auditing is established, including the following steps: The basic attributes of the project are extracted as parameters for the project profile, and the project application materials are deconstructed to form a multimodal representation of the material profile. Define the boundaries of the intelligent audit analysis domain, horizontally covering project progress status, material submission types, and on-site environmental data, and vertically setting analysis depth thresholds, including micro and macro levels; N monitoring indicators are set up at the control points, and M dynamic observation points are set up along the time dimension for each monitoring indicator; P control observation points are set up in the project reference system. By integrating the structured parameters of the project profile, the multimodal features of the material profile, the dynamic observation sequence of N monitoring indicators, and the benchmark data of the P control observation points, a digital profile of the project and a multidimensional feature analysis space are constructed.

9. The digital project management method based on artificial intelligence according to claim 8, characterized in that: The basic attributes of the project include the planned duration range, project type classification label, responsible department identifier, and initial risk level value; The multimodal representation includes semantic embedding vectors of text, numerical matrices of tables, and visual feature maps of images. The micro level includes the time deviation of individual milestone nodes, while the macro level includes the overall progress fluctuation pattern of departmental project groups.

10. A digital project management system based on artificial intelligence, used to implement the management method according to any one of claims 1-9, characterized in that: include: Basic Model Layer: Based on the needs of project lifecycle management, the project entities are digitally abstracted, a digital profile of the project and a multi-dimensional feature analysis framework are constructed, and the analysis domain boundaries of intelligent auditing are established; Computational Grid Layer: Based on the digital profile of the project, an intelligent analysis grid is automatically generated to adapt to multi-source heterogeneous data. The intelligent analysis grid includes a project progress grid and a material feature characterization grid. The project progress data is divided into grids according to the boundary of the analysis domain. The project grid is adaptively partitioned through an unsupervised clustering algorithm, and the grid density parameters are dynamically adjusted for each partition. Intelligent Decision Layer: Based on the digital profile of the project and the intelligent analysis grid, it comprehensively analyzes the risk stress, dynamically determines the critical threshold for risk warning, and outputs risk warning decision management conclusions that include the critical threshold.

Images