A full-cycle engineering project management system and method

Abstract

The present application relates to the technical field of artificial intelligence and engineering project management, and proposes a whole-cycle engineering project management system and method, aiming to solve the problems of data fragmentation, decision-making relying on manual work and low-efficiency of execution collaboration in traditional engineering project management. The system includes a data layer, a processing layer and an application layer: the data layer integrates a historical change data warehouse, dynamic market data, a supply chain collaboration database and a BIM model whole-life-cycle management, realizing multi-source heterogeneous data correlation storage; the processing layer extracts key information of change text and determines priority through an NLP semantic analysis module, and outputs quantitative decision basis combining an intelligent decision support module (risk prediction, multi-source data fusion, BIM simulation), and an automatic execution module realizes optimal task allocation and progress monitoring; the application layer provides an interactive interface to support whole-process visual collaboration. The method includes four stages of change request submission, NLP semantic analysis, approval decision and automatic execution, forming a closed-loop management.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and engineering project management technology, specifically to a full-cycle engineering project management system and method. Background Technology

[0002] In the full lifecycle management of engineering projects, the traditional model faces three major technical challenges that urgently need to be addressed:

[0003] First, data fragmentation is severe, creating information silos. Historical change records, real-time market prices, supply chain data, and BIM models are scattered across independent systems, lacking a unified linkage mechanism. This necessitates manual cross-platform data retrieval during change analysis, which is not only inefficient but also results in incomplete decision-making due to data asynchrony. For example, change cost estimation requires manually querying material prices and historical cases; BIM models and construction change records cannot be automatically linked, making it difficult to quickly assess the impact of changes on the structure.

[0004] Second, the decision-making process relies on manual intervention and has a low level of intelligence. Change requests are mostly unstructured texts, requiring manual extraction of key information, which is prone to omissions or misjudgments; risk assessment relies on personal experience and lacks quantitative analysis tools, making it difficult to provide timely warnings of risks such as project delays and cost overruns; change simulations of BIM models are disconnected from other data and cannot provide multi-dimensional verification for decision-making.

[0005] Third, there is inefficient collaboration and a lack of closed-loop management in the execution process. Task allocation is based on subjective assignment, making it difficult to achieve optimal matching of people and positions; progress monitoring relies on manual reporting, which lags behind actual progress and makes it difficult to detect abnormalities in a timely manner; cross-departmental collaboration lacks a real-time interactive platform, resulting in a disconnect between change execution and decision-making, creating management gaps.

[0006] Therefore, a full-cycle engineering project management system and method are proposed, which forms a closed-loop management through four stages: change request submission, NLP semantic analysis, approval decision-making, and automated execution, thereby solving the above problems. Summary of the Invention

[0007] Technical problems to be solved

[0008] Traditional engineering project management suffers from problems such as data fragmentation, reliance on manual decision-making, and inefficient execution collaboration.

[0009] Technical solution

[0010] To achieve the above-mentioned objectives, the present invention provides the following technical solution: a full-cycle engineering project management system, applied in the field of artificial intelligence and engineering project management technology, the system comprising a data layer, a processing layer and an application layer.

[0011] The data layer is used to integrate multi-source heterogeneous data and provide input data for upper-layer modules, including a historical change data warehouse, a dynamic market data real-time access module, a supply chain collaborative database, and a BIM model full lifecycle management module.

[0012] The processing layer is used to realize data processing and intelligent decision-making, including an NLP semantic analysis module, an intelligent decision support module, and an automated execution module. The NLP semantic analysis module is used to convert unstructured change text into structured data and determine priorities, including a key information extraction engine and a priority intelligent determination system. The intelligent decision support module is used for risk assessment, multi-source data fusion, and BIM change simulation, including a risk prediction engine, a multi-source data fusion platform, and a BIM change simulation subsystem. The automated execution module is used for task generation, allocation, and progress monitoring, including a task allocation and execution system and a progress monitoring and early warning system.

[0013] The application layer provides a user interface, including a change submission interface, an approval decision interface, and an execution monitoring interface, supporting change request submission, approval process visualization, and real-time task progress tracking.

[0014] Preferably, the historical change data warehouse stores complete information on change requests for past projects, including change initiation time, applicant, change type, approval records, cost comparison, schedule impact, and risk events. It uses a relational database to store structured data, and a file system and object storage to store unstructured data. The structured data and unstructured data are linked by a unique identifier.

[0015] Preferably, the dynamic market data real-time access module obtains building material prices, equipment rental fees, and labor cost indices through API interfaces and web crawling technology. After cleaning, these data are stored in a time-series database, supporting price trend queries by time dimension and providing real-time basis for cost estimation.

[0016] Preferably, the key information extraction engine uses a Transformer-based sequence labeling model to parse the changed text and extract key information such as the reason for the change, the content of the change, and the scope of impact. The model captures inter-word dependencies through a multi-head attention mechanism and outputs a label sequence through a CRF layer. The priority intelligent judgment system performs sentiment scoring on the key information text based on an LSTM model and generates priority labels in combination with a preset threshold. High-priority changes automatically trigger an expedited approval process.

[0017] Preferably, the risk prediction engine uses a random forest algorithm to construct a risk assessment model, inputting change text features, project background features, and historical similar change features, and outputting the risk probability distribution of project delays, cost overruns, and safety accidents; the multi-source data fusion platform establishes a weighted fusion model to process four types of data sources: market, supply chain, historical, and BIM, and calculates a comprehensive score to support decision-making; the BIM change simulation subsystem parses change parameters based on the IFC standard, loads the model through the BIM API, and uses the AABB hierarchical bounding box algorithm for collision detection, outputting three-dimensional visualization simulation results.

[0018] Preferably, the task allocation and execution system generates a task list based on a preset change template, uses the Hungarian algorithm to solve the bipartite graph matching problem to achieve optimal task allocation, and the objective function is the comprehensive cost of task allocation, i.e., the lowest time cost and capability matching score; the progress monitoring and early warning system uses the Dynamic Time Warping (DTW) algorithm to match the actual progress with the planned progress, triggers a tiered early warning when the cumulative difference exceeds a threshold, and automatically pushes anomaly handling suggestions.

[0019] A method for managing a full-cycle engineering project, the method being applied to the aforementioned full-cycle engineering project management system, the method comprising the following steps:

[0020] S1. Change Request Submission Stage: Receives change requests from users, supporting voice, text, and image formats. Performs preliminary verification to ensure data integrity and preprocesses the text by word segmentation, stop word removal, and terminology standardization.

[0021] S2. NLP Semantic Analysis Stage: The Transformer model is used to extract key information from the changed text, including the reason for the change, its content, and its scope of impact. The LSTM model is used to score sentiment and determine priority. The extracted results are then encapsulated into a structured change object.

[0022] S3. Approval and Decision-Making Stage: The risk prediction model is activated to output a three-dimensional risk probability matrix. A comprehensive score is calculated through a multi-source data fusion platform. Decision recommendations are generated by combining the results of BIM change simulation, and manual approval and correction are supported.

[0023] S4. Automated Execution Phase: Based on the change type matching preset template, a task list is generated, tasks are assigned using the Hungarian algorithm, repetitive operations are automatically executed through RPA tools, task progress is monitored in real time and anomaly warnings are triggered, forming a closed-loop management system.

[0024] Preferably, the BIM change simulation in step 3 includes: parsing the geometric parameters and attribute parameters in the change content, loading the corresponding version model through BIMAPI and locating the affected components, performing collision detection using the AABB hierarchical bounding box algorithm, and displaying the detection results in a three-dimensional visualization format.

[0025] Preferably, the task list in step 4 includes task nodes, responsible positions, and time requirements, and the dependencies between task nodes are visualized. When the task progress delay exceeds the threshold, the system automatically adjusts the time nodes of subsequent tasks and pushes resource allocation suggestions.

[0026] Preferably, the comprehensive score calculation method for the multi-source data fusion platform in step 3 is as follows:

[0027]

[0028] in, The weights for market data, supply chain data, historical data, and BIM data are respectively. The standardized score range for each data source is 0-100, calculated based on price fluctuations, supply capacity, historical changes, and model modification complexity.

[0029] Beneficial effects

[0030] Compared with the prior art, the present invention provides a full-cycle engineering project management system and method, which has the following beneficial effects:

[0031] 1. In this solution, the full-cycle engineering project management system breaks down information silos through deep integration of multi-source heterogeneous data (historical change data, market data, supply chain data, and BIM models), enabling the correlation and reuse of data throughout the entire lifecycle. Leveraging NLP semantic analysis, risk prediction engines, and intelligent engines for BIM simulation, it transforms unstructured information into quantitative decision-making basis, replacing traditional manual analysis and experience-based judgment, significantly improving the scientific nature and efficiency of decision-making. Through automated execution modules and cross-departmental collaboration interfaces, it achieves closed-loop management of task allocation, progress monitoring, and anomaly handling, greatly reducing manual coordination costs and comprehensively improving the accuracy, collaboration, and intelligence level of full-cycle engineering project management.

[0032] 2. In this solution, the full-cycle project management method uses NLP semantic analysis to achieve structured extraction and intelligent priority determination of change texts, replacing manual sorting and subjective classification, ensuring the consistency and timeliness of information processing; with the help of multi-source data fusion assessment, risk quantification analysis and BIM simulation verification, it provides multi-dimensional scientific basis for approval decisions, avoiding the limitations of traditional experience-based decision-making; through automated task allocation, progress monitoring and anomaly adjustment mechanisms, it achieves precise control and dynamic optimization of the entire change execution process, forming a complete closed loop of "request-analysis-decision-execution", significantly improving the process standardization, decision reliability and execution efficiency of project management. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the overall system architecture of the present invention;

[0034] Figure 2 This is a schematic diagram of the data layer architecture of the present invention;

[0035] Figure 3 This is a schematic diagram of the processing layer architecture of the present invention;

[0036] Figure 4 This is a schematic diagram of the application layer architecture of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see Figures 1-4 This invention proposes a full-cycle engineering project management system and method, including the following:

[0039] I. Full-cycle Engineering Project Management System

[0040] (I) Data Layer: The Integration Hub for Multi-Source Heterogeneous Data

[0041] The data layer, as the foundation of system operation, undertakes the core tasks of data acquisition, cleaning, storage, and management. Its core role is to provide high-quality input data for upper-layer modules. This layer integrates four major categories of data:

[0042] 1. Historical Change Data Warehouse

[0043] This system stores complete information on all change requests from past projects, including the change initiation time, applicant, change type (design / construction / procurement, etc.), change description text, approval process records, implementation data, cost comparison before and after the change, impact on schedule, and risk event records. For example, the historical data of a bridge project contains over 300 pile foundation design change records. Each record is associated with attachments such as geological survey reports, BIM model versions, and construction logs, forming a hybrid storage of structured and unstructured data. The data format uses a relational database (such as PostgreSQL) to store structured data, and a file system or object storage (such as MinIO) to store unstructured data such as drawings and documents, linked by unique identifiers.

[0044] 2. Real-time access to dynamic market data

[0045] Real-time market data such as building material prices (e.g., steel, concrete, pipes), equipment rental costs, and labor cost indices are obtained through API interfaces. For example, steel price fluctuation data is updated every 15 minutes by connecting with commodity trading platforms (e.g., Shanghai Steel Union); quarterly labor wage guidelines are crawled using web scraping technology. After data cleaning, it is stored in a time-series database (e.g., InfluxDB), supporting price trend queries by time dimension, providing real-time basis for cost estimation.

[0046] 3. Supply Chain Collaboration Database

[0047] The system integrates supplier information (qualifications, production capacity, historical cooperation evaluations), inventory data (raw material inventory levels, turnover cycles), and logistics tracking (transportation routes, delivery times, and anomalies). Through integration with the ERP system, it obtains real-time information such as the status of material procurement orders and supplier delivery dates. For example, the supply chain database of a subway project records the concrete supply capacity of over 50 suppliers. When changes lead to an increase in concrete demand, the system automatically assesses the remaining capacity of each supplier (e.g., supplier A's daily production is 300 cubic meters). 3 The current order occupies 200m 3 The minimum delivery time (e.g., supplier B can deliver within 48 hours) provides a quantitative basis for resource allocation.

[0048] 4. BIM model full lifecycle management

[0049] The system stores 3D BIM models of engineering projects, including models for various disciplines such as architecture, structure, and MEP, as well as model version information (e.g., different versions for the design, construction, and completion phases). Each model component is associated with geometric parameters (dimensions, coordinates), attribute parameters (material type, strength grade), and associated documents (design drawings, calculation sheets). For example, the BIM model of a commercial complex contains more than 100,000 components, each with a unique UUID identifier. It supports obtaining detailed information about components (e.g., duct diameter 800mm, material galvanized steel plate) through API interfaces, providing an accurate 3D data foundation for change simulation.

[0050] (II) Processing Layer: The Core Engine of Intelligent Management

[0051] The processing layer is the "brain" of the system, consisting of three core modules. Each module works collaboratively through data interfaces to complete the entire intelligent operation from data processing to decision output.

[0052] 1. NLP Semantic Analysis Module: Intelligent Parser for Changing Text

[0053] The core function of this module is to transform unstructured change request text into structured data, extract key information, and evaluate priorities. Specifically, it includes two sub-functions:

[0054] (1) Key information extraction engine

[0055] A Transformer-based sequence labeling model (such as BERT-CRF) is employed to achieve fine-grained parsing of altered text. The input layer segments the text and converts it into a sequence of word vectors. ,in It is the first Each word is represented by a word embedding (WordPiece) with a dimension of 768 (consistent with the BERT-base model). The model captures inter-word dependencies through a multi-head attention mechanism (12 attention heads). For example, in the sentence "Foundation settlement leads to insufficient pile bearing capacity, requiring an increase of 2 meters in pile length," the causal relationship between "foundation settlement" and "pile bearing capacity" is reflected through attention weights. The output layer uses a CRF layer to predict the label sequence, and the label set is defined as... Each word corresponds to a label, such as "foundation settlement" being labeled as "reason for change" and "increase pile length by 2 meters" being labeled as "content of change".

[0056] During model training, a large-scale labeled dataset (containing 200,000+ historical change texts) was used for fine-tuning, and the loss function was cross-entropy loss.

[0057]

[0058] in, The length of the text (number of words). Number of label categories (4 categories). The real label is 0 or 1, where 1 indicates that the i-th word belongs to the j-th label. The model predicts the probability that the i-th word belongs to the j-th label. Experimental results show that the F1 score for key information extraction can reach over 92%.

[0059] (2) Priority Intelligent Determination System

[0060] An LSTM-based sentiment analysis model assesses the urgency of changes. The input is a text vector concatenated from word vectors containing key information. A three-layer LSTM network (256 hidden units per layer) captures long-range semantic dependencies. (Forget gate) The decision on how much historical information to retain is made through the input gate. Controls the new information currently being input, output gate Regulating cell state Generate hidden state The specific formula is as follows:

[0061]

[0062] in, Indicates a time step. The word vector input at the current time step; , , , These are the weight matrices for the forget gate, input gate, cell state update, and output gate, respectively. , , , For the corresponding bias vector; This is the sigmoid function (output range 0-1). For hyperbolic tangent function (output range -1-1); This represents element-wise multiplication. The LSTM model outputs a sentiment score. A higher score indicates a more urgent change. (Preset threshold) (High priority) (Medium priority), for example, the sentiment score of "the risk of mudslide in the tunnel requires immediate adjustment of the support plan" is 0.8, which is automatically marked as high priority and triggers the expedited approval process.

[0063] 2. Intelligent Decision Support Module: The Decision-Making Hub for Multi-Dimensional Analysis

[0064] This module integrates three major functions: risk assessment, multi-source data fusion, and BIM simulation, providing comprehensive support for approval processes.

[0065] (1) Risk prediction engine

[0066] A risk assessment model is constructed using the random forest algorithm. The input features include:

[0067] Modified text features: Keyword weights extracted via TF-IDF, and sentiment scores of key information.

[0068] Project background characteristics: Project phase (design / construction / operation and maintenance), contract type (EPC / DBB), investment scale

[0069] Characteristics of similar historical changes: frequency of risk occurrence and average duration of impact for similar changes.

[0070] A single decision tree selects the splitting attribute based on Gini impurity, and finally, the prediction results of T=100 decision trees are integrated using a voting method to output the risk probability distribution. These correspond to the probabilities of project delays, cost overruns, and safety accidents, respectively. For example, the predicted results of design changes for a building construction project are: a 35% probability of project delays (threshold 20%) and a 28% probability of cost overruns (threshold 25%), triggering a risk warning.

[0071] (2) Multi-source data fusion platform

[0072] A weighted fusion model is established to process n=4 types of data sources (market, supply chain, history, BIM), and the formula is as follows:

[0073]

[0074] Among them, weight Determined through data importance analysis (such as market prices). Supply chain Historical data BIM simulation Scores from each data source Standardized to the [0, 100] interval:

[0075] Market Score The score is based on the product of the price fluctuation range and the change amount (e.g., a 5% increase in steel prices with a change amount accounting for 10% of the total price, scores 80).

[0076] Supply chain score Based on supply capacity (90% of remaining capacity ≥ demand) and delivery time (90% of delivery time ≤ 24 hours).

[0077] Historical scores Referencing similar change rates (≤10%, 85) and schedule delay rates (≤5 days), 85 is awarded.

[0078] BIM Score Based on the collision detection results (80 for no collisions) and the complexity of model modification (75 for component adjustment amount ≤ 5%), the score is calculated as follows:

[0079] For example, the integration score for a certain drainage pump modification is:

[0080] .

[0081] (3) BIM Change Simulation Subsystem

[0082] The specific steps for implementing the interaction between change information and the BIM model based on the IFC standard are as follows:

[0083] Analyze the geometric parameters (such as component size and position adjustments) and attribute parameters (material replacements) in the changes.

[0084] Load the corresponding version of the model using BIM API (such as Autodesk Forge) to locate the affected component set.

[0085] Collision detection is performed using the AABB hierarchical bounding box algorithm, with the following formula:

[0086]

[0087] in, , The bounding box for the modified component. It is the bounding box of adjacent components. This represents the maximum coordinate of bounding box A on the d-axis (x / y / z). This represents the minimum coordinate of bounding box B on the d-axis. The detection results are displayed in a 3D visualization, with collision areas highlighted in red. For example, if a drain pump (500mm diameter) does not collide with existing pipelines (1500mm spacing), the simulation report will show "sufficient installation space."

[0088] 3. Automated Execution Module

[0089] This module automates the entire process from task generation to progress monitoring, and includes two core functions:

[0090] (1) Task allocation and execution system

[0091] The task list generation is based on a template engine, with 10+ pre-set change templates (such as design change templates and construction plan change templates). Each template defines task nodes (such as drawing modification, budget adjustment, and site briefing), responsible positions (designer, cost engineer, construction manager), and time requirements (drawing approval to be completed within 3 days). Task allocation uses the Hungarian algorithm to solve the bipartite graph matching problem, with the objective function being:

[0092]

[0093] in, For the number of tasks, For the number of responsible persons, The total cost of assigning task i to person j (time cost + ability matching score). Let be the decision variables (1 represents allocation, 0 represents no allocation). The optimal allocation scheme is found using a matrix inversion algorithm. For example, the task of "drawing modification" can be assigned to Engineer Zhang, whose design ability is 90, and the task of "budget adjustment" can be assigned to Engineer Li, whose cost estimation ability is 95.

[0094] (2) Progress monitoring and early warning system

[0095] The Dynamic Time Warping (DTW) algorithm is used to match actual progress with planned progress, with discrete time series as input. and ,in This represents the actual percentage completed. Percentage of planned completion. Distance matrix. The recursive calculation is as follows:

[0096]

[0097] in, For the difference between actual and planned progress (e.g.) ), This is the cumulative distance. When the cumulative distance... When the threshold is set to 0.3, a three-level warning is triggered: yellow warning (delay of 1-3 days), orange warning (delay of 3-7 days), and red warning (delay of >7 days). For example, if Task 3 (purchasing drainage pumps) reports "supplier out of stock" on the second day, the system detects a 50% delay in progress, triggers a yellow warning, and automatically switches to a remote warehouse to transfer goods.

[0098] (III) Application Layer: Unified Entry Point for User Interaction

[0099] The application layer adopts a microservice architecture to develop web and mobile apps, providing three core functional interfaces:

[0100] 1. Change the submission interface

[0101] It supports rich text input, attachment upload (drawings, inspection reports), and BIM model preview. It automatically suggests similar historical change cases to help users improve change descriptions. For example, when you enter "pile foundation", a list of similar changes will pop up, showing the handling solutions and impact data of past pile foundation changes (such as "In 2023, the pile foundation of a certain project was lengthened by 2 meters, increasing the cost by 120,000 yuan").

[0102] 2. Approval Decision Interface

[0103] Visualize the NLP analysis results (key information highlighted), risk assessment reports (probability distribution histograms), and BIM collision detection results (3D model comparison views). Approvers can adjust the weights of various factors through a slider and view the changes in decision-making suggestions in real time, with support for annotations and electronic signatures.

[0104] 3. Execution monitoring interface

[0105] Display the task progress in the form of a Gantt chart, update the completion status of the responsible person in real time (such as "Engineer Zhang - drawing modification: 80% completed"), integrate IoT device data (such as GPS positioning of construction machinery, sensor data), automatically highlight abnormal status in red and display the detailed reasons (such as "Steel not delivered on time - supplier transportation delay"). Visualize task dependencies (such as task 4 starts after task 1 is completed), and intuitively display the process connection.

[0106] II. Full-cycle engineering project management method

[0107] (1) Change request submission stage: conversion from text to structured data

[0108] 1. User input

[0109] The construction worker / designer submits a change request through the mobile APP or the web端, supporting voice input (automatically converted to text) and photo recognition (extracting drawing change marks). For example, the construction worker discovers geological inconsistencies on site, takes a photo of the drill hole columnar section and describes it verbally: "The thickness of the silty clay at K1+200 is 2 meters less than the exploration report, and the pile length needs to be increased."

[0110] 2. Preliminary verification

[0111] The system automatically checks whether the required fields (such as change location, impact scope) are complete and whether the attachment format is compliant (only supports PDF, DWG, IFC). If missing, it prompts the user to supplement (such as "Please fill in the expected impact on the construction period of the change"), ensuring the integrity of the submitted data.

[0112] 3. Text preprocessing

[0113] Perform word segmentation, stop word removal, simplified and traditional Chinese conversion, etc. on the input text to generate a pure word sequence, providing a basis for subsequent NLP analysis. For example, convert "砼" to "concrete", remove meaningless words such as "的" and "了", and unify the terminology expression.

[0114] (2) NLP semantic analysis stage: intelligent extraction and priority determination of key information

[0115] 1. Key information extraction

[0116] The Transformer model analyzes the text word by word, outputting the label probability for each word, and decodes the optimal label sequence using the Viterbi algorithm. For example, given the input text "Due to heavy rain causing water accumulation in the foundation pit, two additional drainage pumps are needed, and the construction period is expected to be extended by 3 days," the extracted result would be:

[0117] Reason for change: Heavy rain caused water accumulation in the foundation pit.

[0118] Changes: Two drainage pumps added.

[0119] Impact: Construction period extended by 3 days

[0120] 2. Sentiment Analysis and Classification

[0121] The LSTM model assigns sentiment scores to key information text and generates priority labels based on preset rules. If the sentiment score S=0.7 (>α=0.6), it is marked as high priority, the system automatically adds an "urgent" label, and notifies the chief engineer via SMS.

[0122] 3. Data encapsulation

[0123] The extracted key information, priorities, original text, attachments, etc. are encapsulated into a JSON format change object, transmitted to the intelligent decision support module through the API interface, and stored in the historical change database, generating a unique change ID (such as BC-20231015-001).

[0124] (III) Approval and Decision-Making Stage: Multi-Dimensional Analysis Supports Scientific Decision-Making

[0125] 1. Risk assessment initiated

[0126] The risk prediction model loads the textual features of the object to be changed, basic project data (such as a contract period of 180 days and a budget of 50 million yuan), and historical similar change data (the average duration impact of 30 drainage system changes in the past year is 2 days), and outputs a three-dimensional risk probability matrix. (Construction period / cost / safety risks).

[0127] 2. Multi-source data fusion computing

[0128] Synchronously call the market data API to obtain the current steel price fluctuation (+5%), and the supply chain system reports sufficient drainage pump inventory (delivery within 24 hours). Combine this with the cost impact of similar changes in historical change data (average cost overrun of 15%) and the complexity of BIM simulation model modifications (component adjustment ≤ 5%), calculate the comprehensive impact score:

[0129]

[0130] 3. BIM Change Simulation

[0131] Analyze the equipment parameters in the changed content (drainage pump dimensions: diameter 500mm, height 1200mm), locate the installation position in the foundation pit model, and detect collisions with existing pipelines (diameter 300mm, spacing 1500mm):

[0132] Lateral spacing: 1500mm ≥ 500mm + 300mm (safety distance), no collision

[0133] Longitudinal height: Pit depth ≥ 5m, ≥ 1200mm (equipment height), ample installation space.

[0134] The simulation report outputs: No collision risk, and the installation space meets the requirements.

[0135] 4. Decision Recommendation Generation

[0136] Based on risk assessment (project schedule risk exceeds threshold), comprehensive score (M=81.25>70), and BIM simulation (no collision), the system generates the suggestion: "Agree to the change, pay attention to the trial operation and testing after the drainage pump is installed", with detailed supporting evidence (risk report, BIM screenshots, market quotation).

[0137] 5. Manual approval and correction

[0138] Approving personnel can browse the system's suggestions and, if they have any questions, can manually trigger a request for supplementary information (such as requesting ground-penetrating radar scan data) or directly modify their decision (such as rejecting a change due to budget constraints). The approval results are fed back to the system in real time.

[0139] (iv) Automated Execution Phase: From Task Allocation to Closed-Loop Management

[0140] 1. Task list generation

[0141] Based on the change type (construction plan change), a preset template is matched to generate a list containing 5 task nodes:

[0142] Task 1: The design department modifies the drainage system drawings (Responsible person: Engineer Zhang, Time limit: 2 days)

[0143] Task 2: Contract Department adjusts budget (Responsible person: Engineer Li, Duration: 1 day)

[0144] Task 3: Procurement of drainage pumps by the Materials Department (Responsible person: Engineer Wang, Time limit: 3 days)

[0145] Task 4: On-site briefing by the Engineering Department (Responsible person: Engineer Zhao, Duration: 0.5 days, dependent on Task 1)

[0146] Task 5: Quality Department Acceptance (Responsible Person: Engineer Zhou, Duration: 0.5 days, dependent on Task 3)

[0147] 2. Intelligent task allocation

[0148] The Hungarian algorithm calculates the optimal allocation scheme based on the competency matrix of the responsible persons (Engineer Zhang: design ability 90, coordination ability 80; Engineer Li: cost ability 95, communication ability 75) to ensure the lowest total cost of the task. For example, the task of revising drawings can be assigned to Engineer Zhang, and the task of adjusting the budget can be assigned to Engineer Li.

[0149] 3. RPA Automated Execution

[0150] Robotic process automation tools automatically perform repetitive tasks:

[0151] Extract the drawing number from the changed object, automatically update the CAD drawing version number, and upload it to the document management system;

[0152] Based on the budget adjustment amount (+50,000 yuan), create a new purchase requisition in the ERP system;

[0153] Generate change notification emails and automatically copy all relevant parties (owner, supervisor, subcontractor).

[0154] 4. Real-time progress monitoring

[0155] The system synchronizes the status of each task every hour:

[0156] Task 1: At 16:00 on day 1, Engineer Zhang marks "Drawing modification completed," and the system automatically triggers the start condition for Task 4.

[0157] Task 3: On the second day, Engineer Wang reported that "the supplier notified that the drainage pump is out of stock." The system detected a 50% delay in progress (66% planned, 16% actual), triggering a yellow alert and prompting the materials department to initiate the alternative supplier process (automatically switching to a remote warehouse for restocking, with an estimated delivery delay of 1 day).

[0158] 5. Exception handling and closed-loop management

[0159] In response to the schedule delay, the project manager initiated a coordination meeting on the monitoring interface, adjusting the procurement plan to "transfer goods from a remote warehouse, with an estimated delivery delay of 1 day." The system automatically updated the duration of Task 3 (3 days → 4 days) and recalculated the time nodes for subsequent tasks (the start time of Task 5 was delayed by 1 day). After the change was implemented, the quality department uploaded an acceptance report (100% pass rate), and the system automatically archived all relevant data (change requests, approval records, execution logs), forming a complete change management closed loop.

[0160] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating the present invention and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A whole-cycle engineering project management system applied to the field of artificial intelligence and engineering project management, characterized in that: The system comprises a data layer, a processing layer and an application layer; The data layer is used for integrating multi-source heterogeneous data and providing input data for upper layer modules, comprising a historical change data warehouse, a dynamic market data real-time access module, a supply chain collaboration database and a BIM model full life cycle management module; the historical change data warehouse uses a relational database to store structured data, uses a file system or object storage to store unstructured data, and associates structured data with unstructured data through a unique identifier; The processing layer is used for realizing data processing and intelligent decision-making, comprising an NLP semantic analysis module, an intelligent decision support module and an automatic execution module; the NLP semantic analysis module is used for converting unstructured change text into structured data and determining priority, comprising a key information extraction engine and a priority intelligent judgment system; the key information extraction engine uses a sequence labeling model based on Transformer to extract change reason, change content and impact range; the priority intelligent judgment system performs sentiment scoring on key information based on an LSTM model and generates a priority label; The intelligent decision support module is used for risk assessment, multi-source data fusion and BIM change simulation, comprising a risk prediction engine, a multi-source data fusion platform and a BIM change simulation subsystem; the risk prediction engine uses a random forest algorithm to output risk probability distribution of schedule delay, cost overrun and safety accidents; the multi-source data fusion platform establishes a weighted fusion model to process market, supply chain, historical and BIM data sources and calculates a comprehensive score; the BIM change simulation subsystem analyzes change parameters based on IFC standard and performs collision detection using an AABB hierarchical bounding box algorithm; The automatic execution module is used for task generation, distribution and progress monitoring, comprising a task distribution and execution system and a progress monitoring and early warning system; the task distribution and execution system uses the Hungarian algorithm to realize optimal task distribution, with a target function of minimizing time cost and capability matching degree score; the progress monitoring and early warning system uses a dynamic time warping algorithm to match actual progress with planned progress and triggers hierarchical early warning when cumulative difference exceeds a threshold value; The application layer is used for providing a user interaction interface, comprising a change submission interface, an approval decision interface and an execution monitoring interface, supporting change request submission, approval process visualization and real-time tracking of task progress.

2. A full-cycle project management system according to claim 1, characterized in that: The dynamic market data real-time access module obtains building material prices, equipment rental fees and labor cost indexes through API interfaces and crawler technology; after data cleaning, the data is stored in a time series database, supporting query of price trends according to time dimension.

3. The full-cycle engineering project management system of claim 1, wherein: The key information extraction engine uses a sequence labeling model to capture inter-word dependency relationships through a multi-head attention mechanism, and outputs a label sequence through a CRF layer.

4. The full-cycle engineering project management system of claim 1, wherein: In the multi-source data fusion platform, the weights of market data, supply chain data, historical data and BIM data are dynamically adjusted or preset based on data importance analysis.

5. The full-cycle engineering project management system of claim 1, wherein: In the task allocation and execution system, the time cost in the objective function is calculated based on the task estimated duration and the current load of the responsible person, and the ability matching degree score is calculated based on the historical task completion quality and skill label matching degree of the responsible person.

6. The full-cycle engineering project management system of claim 1, wherein: The progress monitoring and early warning system triggers yellow, orange or red three-level early warning according to the size of the cumulative difference value, and automatically pushes the corresponding abnormal treatment suggestions.

7. A full-cycle project management method, characterized by: The method is applied to the whole-cycle engineering project management system of any one of claims 1-6, and the method comprises the following steps: S1. Change request submission stage: receiving user input change request, preliminary verification and text preprocessing; S2. NLP semantic analysis stage: using the Transformer model to extract key information change reasons, content, impact scope in the change text, performing sentiment scoring and priority determination through the LSTM model, and encapsulating the extraction results as structured change objects; S3. Approval decision stage: starting the risk prediction model to output a three-dimensional risk probability matrix, calculating a comprehensive score through a multi-source data fusion platform, generating decision suggestions combined with BIM change simulation results, supporting manual approval and correction; wherein the BIM change simulation includes analyzing geometric parameters and attribute parameters in the change content, and performing collision detection using the AABB hierarchical bounding box algorithm; the comprehensive score calculation method of the multi-source data fusion platform is: wherein, are the weights of market data, supply chain data, historical data, BIM data, respectively, are the standardized scores of each data source, respectively; S4. Automatic execution stage: generating a task list based on change type matching a preset template, assigning tasks using the Hungarian algorithm, automatically executing repetitive operations through RPA tools, monitoring task progress in real time and triggering abnormal early warning, and forming a closed-loop management.

8. A whole-cycle project management method according to claim 7, characterized in that: The task list is visualized in the form of a Gantt chart to show task nodes, responsible positions, duration requirements and the dependency relationship between tasks, and when the task progress delay exceeds a threshold value, the system automatically adjusts the time nodes of subsequent tasks and pushes resource allocation suggestions.

Images