Intelligent project management method and system for project supervision

Abstract

The invention provides an intelligent project management method and system for project supervision, and relates to the technical field of project supervision, and the method comprises the steps: taking a structural member and a construction technology in a construction drawing as a basis, extracting a risk supervision unit with a space anchor point and a supervision priority through a semantic matching and structural modeling mode, and carrying out the construction of a project. The risk units serve as the minimum execution entity of the supervision task, the future disturbance trend of each risk unit is predicted based on the time sequence modeling technology in combination with field IoT data and the construction progress, and the dynamic risk score of the risk unit is calculated according to the future disturbance trend; by using the scoring system, through a matching algorithm and a scheduling optimization model, dynamically generating an optimal scheduling suggestion of the supervisor, and continuously recording supervision behaviors of the supervisor on site; all supervision behaviors are subjected to semantic alignment with drawing specifications in a structured filing mode, the behavior quality is judged through a consistency scoring mechanism, and acceptance suggestions or rectification actions are triggered on the basis; and the consistency of the supervised object and the drawing target is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering supervision, and in particular to an intelligent project management method and system for engineering supervision. Background Art

[0002] Current project supervision is facing unprecedented challenges. With the increasing scale and complexity of projects, traditional plan-centric supervision methods that rely on manual inspections are struggling to meet the dual requirements of quality and schedule in the face of high parallelism, frequent changes, and complex operating environments. Critical construction nodes are often missed due to disjointed plans or insufficient staffing, leading to potential risks not being identified in a timely manner. Furthermore, while an increasing number of projects are introducing smart site systems equipped with technologies such as video surveillance, IoT sensors, and personnel location systems, most of these systems lack connectivity and collaboration. Data collected onsite remains at the display level, difficult to translate into triggers for supervisory actions and support decision-making. More critically, construction drawings, the core basis of engineering projects, have not systematically extracted and modeled the structural intent, process specifications, and construction cadence they contain. This results in a lack of a clear anchor point and a basis for dynamic adjustment of supervision objectives. In addition, during the project completion acceptance stage, manually filled-out paper acceptance forms and photo and video evidence are still the main means of verification. There is a lack of a comprehensive verification mechanism that connects supervision behavior, drawing standards and construction site status, making it impossible to achieve high-credibility process review and responsibility tracing.

[0003] Therefore, how to extract operational supervision anchor points from source drawings, combine them with dynamic disturbance prediction of the construction process, drive supervision scheduling and behavior archiving, and ultimately achieve consistency verification and rectification feedback of full-process supervision has become a key technical issue that needs to be solved urgently. Summary of the Invention

[0004] To solve the above problems, the present invention provides an intelligent supervision method and system driven by drawing semantics, which constructs a full closed-loop system from design understanding, disturbance prediction, dynamic scheduling to supervision filing and acceptance control.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] In a first aspect of the present invention, there is provided an intelligent project management method for engineering supervision, comprising the following steps:

[0007] S1. Analyze the 2D construction drawings, 3D structural models, and design documentation in traditional engineering projects into a set of risk supervision units with engineering structure, construction intent, timing constraints, and supervision priorities.

[0008] S2. Based on the extracted risk supervision unit set data, evaluate the potential risk change trend of each risk node and obtain a dynamic risk score set by calculating the risk intensity value of each risk node;

[0009] S3. Based on the risk intensity score of each risk node, calculate the matching degree between each risk node and each supervisor, and dynamically adjust the supervisor scheduling plan based on the matching degree score between each supervisor and the node;

[0010] S4. The supervisor schedule and the supervisor behavior records during the actual construction process are structured and archived. During the actual construction process, the system will automatically mark any abnormal supervisory behavior and data;

[0011] S5. Build a supervision consistency model to verify the consistency between supervision behavior, construction records, and drawing standards, and output the consistency score of each node's supervision behavior and drawing standards and the total consistency confidence of each construction node;

[0012] S6. Generate an acceptance rectification strategy based on the output results of the supervision consistency model, and convert the acceptance rectification strategy into a specific scheduling execution item.

[0013] Preferably, in said S1, in said risk supervision unit set, each risk supervision unit includes component type and number, process description text, structural connectivity and complexity score, priority score and belonging process and planned time.

[0014] Preferably, in S2, the risk intensity value of each risk node is calculated by the following formula:

[0015]

[0016] in, Represents the risk node R i The dynamic risk score of , α represents the weighting coefficient, Represents node R i The predicted value of the disturbance at the future time t+Δt, ρ i For node R i risk priority score.

[0017] Preferably, in S3, the matching degree between each risk node and each supervisor is calculated to indicate whether the supervisor is suitable for supervising the node. The higher the matching degree score, the more suitable the supervisor is for supervising the node.

[0018] The supervisory scheduling plan prioritizes high-risk nodes for supervision, and the constraints include:

[0019] Each node must have at least one supervisor to supervise;

[0020] The working hours of each supervisor cannot exceed their maximum available time;

[0021] Each node must complete supervision within the specified time.

[0022] Preferably, in S4, whenever the supervisor arrives at a construction node and starts supervision, a supervision behavior record is automatically generated, which includes the following information:

[0023] Action type, including inspection, acceptance, and corrective action recommendations, to indicate the nature of the action;

[0024] Supervisor number, indicating the person who performs the supervisory task;

[0025] Construction node number, indicating the construction node corresponding to this supervision behavior;

[0026] Timestamp, indicating the start and end time of the recorded supervision behavior;

[0027] Behavior content refers to the specific description of supervisory behavior, including acceptance reports and inspections for compliance with construction specifications.

[0028] Preferably, in S5, the loss function of the supervised consistency model is as follows:

[0029]

[0030] in, represents the loss function, G represents the Euclidean distance between the drawing specification and the supervision record in the semantic space, which is used to measure the consistency difference; i Represents a structured vector; Represents the supervised semantic vector; represents the regularization term, which is used to encourage the model to avoid overfitting to redundant fields; β is the weight of the regularization term.

[0031] Preferably, in S6, an acceptance priority score is constructed for each construction node, and the calculation method is as follows:

[0032]

[0033] Among them, Π i Indicates the acceptance priority score; represents the node supervision consistency score; r i Indicates the node construction risk level; Indicates the node's specified completion time; t now Indicates the current system time; β1, β2, β3 indicate the preset scheduling strategy weights;

[0034] The priority score is used to sort nodes and decide whether each node should be automatically accepted or trigger rectification suggestions. Two threshold values ​​are set: acceptance threshold θ accept and rectification threshold θ fix ;

[0035] like And Π i If the value is less than the preset threshold, an action suggestion that passes the acceptance test will be generated directly;

[0036] like And Π i >θ fix , then generate rectification suggestions and assign rectification execution personnel;

[0037] Otherwise, the node status remains pending.

[0038] In a second aspect of the present invention, there is also provided an intelligent project management system for engineering supervision, comprising:

[0039] The data processing module is used to parse the 2D construction drawings, 3D structural models and design description documents in traditional engineering projects into a set of risk supervision units;

[0040] The risk assessment module is used to evaluate the potential risk change trend of each risk node based on the extracted risk supervision unit set data, and obtain a dynamic risk score set by predicting the risk intensity value of each risk node;

[0041] The risk matching scheduling module is used to calculate the matching degree between each risk node and each supervisor based on the risk intensity score of each risk node, and dynamically adjust the supervisor scheduling plan based on the matching degree score between each supervisor and the node;

[0042] The abnormal data marking and data storage module is used to structure and archive the supervisory schedule and supervisory behavior records during the actual construction process. During the actual construction process, the system will automatically mark any abnormal supervisory behavior and data;

[0043] Supervision consistency module, used to build a supervision consistency model to verify the consistency between supervision behavior, construction records and drawing standards;

[0044] The acceptance and rectification module is used to generate an acceptance and rectification strategy based on the output results of the supervision consistency model, and convert the acceptance and rectification strategy into specific scheduling execution items.

[0045] The beneficial effects of the present invention lie in: addressing the current problems of disconnected sources, uncontrolled processes, and unverifiable results in project supervision processes, this invention proposes an intelligent supervision method and system driven by drawing semantics. This system establishes a fully closed-loop system, from design understanding, disturbance prediction, dynamic scheduling, to supervision documentation and acceptance control. Based on the structural components and construction processes in construction drawings, this method uses semantic matching and structural modeling to extract risk supervision units with spatial anchor points and supervision priorities, serving as the minimum execution entity for supervision tasks. The system combines on-site IoT data with construction progress, predicts the future disturbance trend of each risk unit using time series modeling techniques, and calculates a dynamic risk score based on this. Leveraging this scoring system, the system dynamically generates optimal scheduling recommendations for supervisors through a matching algorithm and a scheduling optimization model, and continuously records their on-site supervision activities. All supervisory activities are semantically aligned with the drawing specifications through a structured documentation process. The quality of these activities is assessed through a consistency scoring mechanism, which then triggers acceptance recommendations or corrective actions. The entire system ensures the consistency between the supervision object and the drawing objectives, the matching of behavior execution and risk level, and the traceability of the acceptance link, realizing a new intelligent supervision method with design goals as the core, disturbance evolution as the driving force, and supervision behavior as the closed loop. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The present invention is a flow chart of an intelligent project management method for engineering supervision.

[0047] Figure 2 This is a framework diagram of an intelligent project management system for engineering supervision according to the present invention. DETAILED DESCRIPTION

[0048] See also Figure 1 As shown, in a first aspect of the present invention, an intelligent project management method for engineering supervision is provided, comprising the following steps:

[0049] S1. Analyze the 2D construction drawings, 3D structural models, and design documentation in traditional engineering projects into a set of risk supervision units with engineering structure, construction intent, timing constraints, and supervision priorities.

[0050] The goal of this step is to parse the two-dimensional construction drawings or three-dimensional structural models (such as BIM) and design description documents in traditional engineering projects into a set of "risk supervision units" with engineering structure, construction intentions, timing constraints and supervision priorities.

[0051] These supervision units {R iThe project management system serves as the unified anchor for subsequent system operations, including scheduling, behavior collection, condition monitoring, and risk assessment. Unlike traditional project management practices that separate structural information, construction tasks, and supervision tasks, this solution integrates these three types of information into a unified structural entity for the first time.

[0052] The original input consists of three parts:

[0053] 2D CAD drawings or 3D BIM model files (IFC format), delivered by the design company and generally uploaded through the engineering project management platform;

[0054] Design description documents, usually in PDF or Word format, containing the node construction sequence, material usage instructions, and special processes (such as formwork requirements and maintenance cycle);

[0055] A task schedule, such as a Gantt chart or a process plan exported from P6, is used to extract the estimated start and end times for each node;

[0056] These data are archived uniformly by the engineering contracting unit at the beginning of the project and can be directly input into the system.

[0057] First, it is necessary to identify the structural components from the drawings or models. For 2D CAD files, contour detection and object layer recognition can automatically extract information such as element boundaries, component types, and annotation text. Taking AutoCAD as an example, the system reads the element attribute layer (such as A-WALL, A-BEAM, etc.), combines the boundary dimensions and annotation text, and identifies the components as columns, beams, walls, floor slabs, etc., and stores them in the preliminary component set {C i For 3D BIM models, directly parse various IfcBuildingElements in the IFC file, such as IfcColumn and IfcBeam, which contain position parameters (x i ,y i ,z i ), dimensions, material properties, associated component ID, etc.

[0058] For example, three columns in a building's main structure are identified, with the drawing annotations "C1, C2, C3." These columns correspond to three IfcColumn components in IFC, with positions of (12.4, 8.1, 0.0), (12.4, 12.1, 0.0), and (12.4, 16.1, 0.0), respectively.

[0059] The next step is to semantically associate the components with the design description text. The text is usually a natural language description written by engineers, such as: "All load-bearing columns must be positioned and reinforced within three days after the ground beam is poured", "The construction of the post-cast strip should be postponed for 21 days and no loading should be allowed during this period." After each description text is divided into sentences, it is sent to a dedicated engineering semantic encoding model (for example, a language model based on BERT fine-tuned and with an engineering terminology vocabulary) to obtain the embedding vector Embed(T j ). The component annotation text in the drawing (such as "load-bearing column C1") is also coded as Embed (C i ), and calculate the cosine similarity between the two:

[0060]

[0061] Among them, r ij ∈[0,1] represents component C i and design description paragraph T j The semantic matching degree of Embed(C i ) represents component C i Annotation word vector of Embed(T j ) indicates the regulation paragraph T j The semantic vector of ; cos represents the cosine similarity function.

[0062] The most matching paragraph T j Bound to the construction constraint semantics of this component.

[0063] After the matching is completed, all components are given process semantic information and their structural positions are marked. i Contains four properties: type, position, matching text, and drawing number.

[0064] Next, priority scoring is performed. Given the limited supervision resources, it is necessary to select the most critical risk points and set the focus of supervision. A combined scoring method combining "structural connectivity" and "process complexity" is proposed.

[0065] ConnectivityConn(C i ) is realized by the number of connections between components in the structural topology, such as the number of connections between columns and beams, and between beams and plates. Complexity Comp(T j ) is measured by the depth of its matching text syntax tree or the number of constraints it contains.

[0066] The composite score is calculated as follows:

[0067] ρ i =λ1·Conn(C i )+λ2·Comp(T j ),

[0068] Among them, ρ i It is component C i The supervision priority score of Conn(C i ) represents component C i Structural connectivity of Comp(T j ) indicates the matching construction instruction paragraph T j Process complexity score.

[0069] For example, if a component connects four beams, and its process specification contains three structural sentences, then:

[0070] ρ i =0.6×4+0.4×3=3.6,

[0071] Finally, component C i Mapped to risk supervision unit R i , the structure contains:

[0072] Component position: 3D coordinates (x i ,y i ,z i ), read from CAD entities or IFC models;

[0073] Component type and number: extracted from the drawing annotation layer;

[0074] Process description text: comes from the design specification and is bound through semantic matching;

[0075] Structural connectivity and complexity scores: derived from topological graphs and grammatical analysis;

[0076] Risk priority score ρ i : As a reference for supervised sorting;

[0077] The corresponding process and planned time: calibrated by Gantt chart.

[0078] The output of this step is a set of structured risk supervision units {R i}, each R i It includes structural identification, spatial anchor points, process semantic constraints, supervision priorities, etc., and is the object basis for subsequent dynamic monitoring, scheduling and consistency verification.

[0079] S2. Based on the extracted risk supervision unit set data, evaluate the potential risk change trend of each risk node and obtain a dynamic risk score set by calculating the risk intensity value of each risk node;

[0080] In this step, the potential risk change trend of each risk node is evaluated by real-time monitoring of the construction site status and combining the risk supervision unit data extracted in step one.

[0081] During the construction process, considering that changes in on-site conditions (such as temperature and humidity, population density, construction progress, etc.) may lead to deviations from the originally designed construction progress, quality, and structural safety, the present invention dynamically predicts these risk fluctuations to promptly identify potential construction problems, ensure that supervisors can prioritize supervision in high-risk areas, and provide effective support for subsequent scheduling decisions.

[0082] The data mainly comes from the output of step S1, that is, the risk node set {R i}, including the following:

[0083] Risk node attributes: including spatial location, component type, process constraints, risk priority ρ i wait;

[0084] Real-time on-site data: Sensor data from the construction site (such as temperature and humidity, meteorological conditions, equipment operating status, personnel distribution, construction progress, etc.) is collected in real time through on-site Internet of Things (IoT) devices, sensors, video surveillance, and personnel positioning systems.

[0085] In actual operation, this step includes:

[0086] On-site status data collection and preprocessing:

[0087] First, construction site status data is collected in real time via IoT sensors (such as temperature sensors, humidity sensors, vibration sensors, and equipment operation sensors). This data is typically recorded as a continuous time series, with each piece of data containing a timestamp and corresponding value (for example, a temperature of 25°C or a humidity of 60%). The collected raw data first needs to be denoised and standardized to remove abnormal data (such as extreme values ​​caused by equipment failure). Different data sources are then normalized to bring all data to the same scale.

[0088] For example, temperature data may have a range (e.g., 0°C to 50°C), while humidity data may have a range (0% to 100%). To enable both types of data to be used together in the model, the present invention converts them into a standard range (e.g., 0 to 1) through normalization. Furthermore, the data is aligned based on timestamps to ensure that data from different sensors are consistent in time, thus ensuring that the input data of the prediction model is synchronized.

[0089] Construction of disturbance prediction model based on historical data:

[0090] By combining the real-time status data of the construction site with the risk node data extracted in step 1, the present invention establishes a time series-based deep learning model (LSTM) to predict the risk of each risk node R i future disturbance trends.

[0091] Specifically, the model is used to predict the status changes or risk fluctuations of a risk node in the future based on historical status data (such as temperature, humidity, population density, etc.).

[0092] Assume that at time t, the historical state data of the node {X i (t-1),X i (t-2),…,X i (tn)}, where X i (tk) represents node R i State data at the past k moments (such as temperature, humidity, construction progress, etc.). Based on these historical state data, the LSTM network is trained through the back propagation algorithm to generate the predicted value of the node's future disturbance That is, the prediction node R i The intensity of risk disturbance at future time t+Δt.

[0093] The prediction model of this process can be expressed as:

[0094]

[0095] in, Represents node R i The predicted value of the disturbance at the future time t+Δt; X i (tk) is the node R i The state data at the past k moments; θ is the training parameter of the LSTM model. By training on historical data, the model can learn the dynamic changes of each node and generate future disturbance predictions for each node.

[0096] Dynamic risk scoring of risk nodes:

[0097] Using the predicted disturbance value By calculating a dynamic risk score for each risk node This score represents the potential risk of the node at the future time t+Δt.

[0098] The calculation of dynamic risk score takes into account the disturbance intensity and the priority of risk node ρ i ,The higher the priority of the node, the more risk fluctuation it should receive more monitoring resources.

[0099] This paper proposes a weighted scoring method that combines the disturbance prediction value and the risk node priority:

[0100]

[0101] in, is the predicted future disturbance value, ρ i is the risk priority score defined in step S1, and α is the weighting coefficient that controls the relative importance of the disturbance value and the priority score in the risk score. Through this formula, the actual disturbance of the node is taken into account with its risk priority to generate a dynamic risk score. Used for subsequent scheduling and monitoring decisions.

[0102] For example, if node R1 has a higher risk priority (e.g., ρ1=0.8), and its future disturbance prediction value If it is 0.6, then its dynamic risk score is:

[0103]

[0104] The main outputs of this step are:

[0105] Dynamic risk score aggregation Each risk node R i are assigned a predicted risk intensity value Indicates its potential future risk fluctuations.

[0106] These dynamic risk scores will serve as the basis for generating subsequent scheduling recommendations, helping the system prioritize scheduling high-risk nodes and ensuring that supervision resources are prioritized in high-risk areas.

[0107] S3. Based on the risk intensity score of each risk node, calculate the matching degree between each risk node and each supervisor, and dynamically adjust the supervisor scheduling plan based on the matching degree score between each supervisor and the node;

[0108] During the construction supervision process, the proper scheduling of supervision resources is key to ensuring construction quality and progress. The risk prediction in step S2 provides a dynamic risk score for each risk node. Based on these scores, efficient supervision scheduling recommendations are generated.

[0109] Step S3 aims to calculate the risk intensity of each node (i.e. the predicted disturbance value) ) and the work situation of the supervisors, dynamically adjust the supervision scheduling plan, and prioritize the allocation of supervision resources to high-risk nodes to achieve the best supervision efficiency and risk control effect.

[0110] This step not only relies on the dynamic risk score of the risk node, but also takes into account the idle time, workload and specific skills of the supervisor to ensure that each node is supervised by the most suitable supervisor. To achieve this goal, this step introduces a risk-driven scheduling suggestion generation method, which combines optimization algorithms and multi-factor weighted scoring to dynamically adjust the task allocation of supervisors.

[0111] The input of this step comes from the output of step S2, i.e. the set of dynamic risk scores Each risk node R i The disturbance intensity prediction value at future time t+Δt Reflects the risk change trend of the node, and the system will schedule according to this data. At the same time, the input also includes the resource information of the supervisors (such as available time, professional skills, etc.) and the construction progress information (such as the construction period and task priority of each node, etc.).

[0112] The specific operation is as follows:

[0113] S301, supervisor resource modeling:

[0114] When building the supervisor resource model, consider the available time T j and professional skills Skill(M j ) of each supervisor. For each supervisor M j , the available time can be automatically obtained through the construction site scheduling system, usually based on the construction site calendar and the schedule. The skill of the supervisor Skill(M j ) is quantified by analyzing its historical work experience, professional certification, and expertise, forming a skill evaluation value that represents its ability to handle certain supervision tasks.

[0115] For example, if a supervisor is good at concrete pouring work, then in nodes related to concrete construction (such as R1), the skill score Skill(M j ) will be higher, while in nodes related to electrical construction, the skill score will be lower.

[0116] S302, calculation of node risk and personnel matching:

[0117] To effectively schedule, first calculate the matching degree between each risk node and each supervisor. This invention calculates the priority of each node and the matching degree of the supervisor through a weighted method. The matching degree of the supervisor and the node is not only related to the risk intensity of the node, but also closely related to the skills and idle time of the supervisor. The matching degree score M ij The calculation formula is as follows:

[0118]

[0119] in: is node R i Risk score of Skill(M j ,R i ) is the supervisor M j For node R i Skill matching degree, the value range is 0 to 1; TimeAvailable(M j ) is the supervisor M j The available time indicates whether it has enough time to access node R in a specific time period. i Monitor; α1, α2, α3 are weight coefficients, which respectively represent the relative importance of risk intensity, skill matching and available time in the matching score.

[0120] This formula calculates a match score for each supervisor and each node, indicating whether the supervisor is suitable for supervising the node. The higher the match score, the more suitable the supervisor is for the node.

[0121] S303, Optimization Scheduling Algorithm:

[0122] Based on the matching score M between each supervisor and the node ij , using integer linear programming (ILP) to generate optimal supervisory schedule recommendations. The goal is to maximize supervisory coverage of high-risk nodes while ensuring that supervisors' workloads do not exceed their available time. It is hoped that each risky node will receive appropriate supervisory personnel during its high-risk period, while ensuring that supervisors' tasks are appropriately scheduled and their workloads are balanced.

[0123] Assuming there are n risk nodes and m supervisors, by introducing binary decision variables X ij , indicating that the supervisor M j Is it assigned to node R? i The objective function is:

[0124]

[0125] Among them, X ij =1 means supervisor M j Assigned to node R i , otherwise X ij = 0. The objective function Z maximizes the sum of the matching scores, that is, giving priority to high-risk nodes for supervision and balancing the workload of supervisors as much as possible.

[0126] Constraints include:

[0127] Each node must have at least one supervisor to supervise;

[0128] The working hours of each supervisor cannot exceed their maximum available time;

[0129] Each node must complete supervision within the specified time.

[0130] The main outputs of this step are:

[0131] Supervisor scheduling suggestions j ,R i ,t}, that is, each supervisor M j Assigned to which risk nodes R i , and determine the specific supervision time t.

[0132] These scheduling suggestions will be passed directly to the supervisors on the construction site to ensure that high-risk nodes are supervised first and the supervisors' workload is balanced.

[0133] S4. The supervisor schedule and the supervisor behavior records during the actual construction process are structured and archived. During the actual construction process, the system will automatically mark any abnormal supervisory behavior and data;

[0134] The core task of this step is to convert the supervisor scheduling suggestion data {M j ,R i ,t}(supervision personnel, construction nodes and supervision time) and supervision behavior records during the actual construction process are structured and archived.

[0135] By associating the supervisor's behavior at each construction node with the corresponding timestamp, supervision content and other information, and storing it as structured data, this step provides basic data for subsequent supervision consistency verification, quality assessment and responsibility tracing.

[0136] Recording and marking of supervisory actions:

[0137] In this step, the behavior of each supervisor (such as inspection, acceptance, supervision, etc.) will be based on the scheduling suggestions in step 3 {M j ,R i ,t} to record and mark. Whenever the supervisor M j Arrival at construction node R i When supervision begins, the system will generate a supervision record. The record includes the following information:

[0138] Action type: such as "inspection", "acceptance" or "corrective suggestions", indicating the nature of the action;

[0139] Supervisor No. M j : Personnel who perform supervisory tasks;

[0140] Construction node number Ri : The construction node corresponding to the current supervision behavior;

[0141] Timestamp T ij : Records the start and end time of the supervision behavior;

[0142] Behavior content: Detailed description of the supervision behavior, such as checking whether it meets the construction specifications, acceptance report, etc.

[0143] This data will be stored in the form of a data packet and associated with the supervision scheduling suggestion data to form a structured record. The format of each supervision behavior data packet includes: ij = {M j , R i , t, B ij , T ij , D ij},

[0144] Where: E ij represents the supervision behavior data packet performed by the supervisor M j on the construction node R i at time t; B ij is the type of supervision behavior (such as "inspection", "acceptance", etc.); T ij is the timestamp, recording the start and end time of the supervision behavior; D ij is the detailed description of the supervision behavior content (such as "check whether it meets the specifications", "acceptance qualified", etc.).

[0145] For example, if the supervisor M1 performs acceptance work on the construction node R1 (such as steel reinforcement welding process), the recorded behavior data packet may include:

[0146] Supervision behavior type: acceptance;

[0147] Supervision content: Check if the steel reinforcement welding is qualified;

[0148] Timestamp: 2023-06-1010:00:00.

[0149] Behavior record data storage and structured management

[0150] All supervision behavior records will be stored in the database to ensure that subsequent searches and management can be performed according to the construction node, supervisor, behavior type, time, etc. The data recorded in the database will include:

[0151] Node_ID: Construction node identifier, indicating the specific construction task or location;

[0152] Supervisor_ID: Supervisor identifier, indicating the personnel performing the supervision behavior;

[0153] Behavior_Type: supervision behavior type, such as "inspection", "acceptance", and "correction suggestions";

[0154] Timestamp: the timestamp of the monitoring behavior;

[0155] Behavior_Content: Description of the content of the supervisory behavior, such as "check whether it complies with the construction specifications";

[0156] Data: Other relevant data, such as attachments, photos, videos, etc.

[0157] Through this structured storage, the system can efficiently query, retrieve, and trace each supervisory action. Each action record can quickly find the specific data associated with it when needed, ensuring data integrity and traceability.

[0158] Abnormal behavior marking and reporting:

[0159] During this step, the system automatically flags any unusual supervisory behavior or data. For example, if a supervisor fails to arrive at a high-risk node for supervision as planned, or fails to conduct acceptance inspections according to regulations, the system will automatically mark it as "supervision failure" and generate a report for subsequent review.

[0160] If a supervisor fails to arrive at a construction node at the scheduled time or fails to inspect a construction node as required, the system will automatically record it as "missing supervision" and generate an exception data report. The report details the time of the incident, the missed supervision node, the responsible supervisor, and the reason, providing a basis for subsequent accountability and decision-making.

[0161] For example, if supervisor M2 fails to arrive at node R3 to inspect the work within the specified time, the system will mark this behavior as "lack of supervision" and generate the following record:

[0162] Type of supervisory behavior: missing;

[0163] Supervision content: Failure to arrive at node R3 for acceptance on time;

[0164] Timestamp: 2023-06-10 08:30:00;

[0165] Note: The supervisor did not arrive at the scheduled time.

[0166] The main outputs of this step are:

[0167] 1. Supervision behavior and construction data package {E ij}, i.e., structured archiving of each supervisory action and construction data;

[0168] 2. Abnormal data report, indicating any supervision deficiencies or abnormal situations, for subsequent responsibility tracing and decision-making.

[0169] S5. Build a supervision consistency model to verify the consistency between supervision behavior, construction records, and drawing standards, and output the consistency score of each node's supervision behavior and drawing standards and the total consistency confidence of each construction node;

[0170] This step aims to construct a comprehensive verification model for consistency verification between supervisory activities, construction records, and drawing standards. The core goal is to quantify the coverage of supervisory activities and the degree of construction execution standardization in a structured manner, providing actionable verification indicators for subsequent risk control and decision-making. Unlike traditional static comparison methods, this approach leverages the rich behavioral semantic information in supervisory data and constructs a characterization vector for drawing standards, achieving vector space mapping and consistency modeling between supervisory and drawing data.

[0171] The input is the set of supervision behavior and construction data packets output in step S4: {E ij}={M j ,R i ,t,B ij ,T ij ,D ij}, where each E ij Indicates supervisor M j At time t, the construction node R i A supervision action performed, including supervision type B ij , time range T ij And supervised text description D ij The structural elements and texts in these data packages will all be used in the modeling process in this step.

[0172] Specific details of the steps:

[0173] Construction of drawing specification information coding module

[0174] All construction nodes R i The corresponding technical specifications of the drawings (such as "the main reinforcement diameter is 16mm, the spacing is 200mm, and the concrete grade is C30") are parsed by a pre-prepared natural language parser to extract structured fields such as steel bar size, quantity, concrete strength grade, etc., and convert this information into a structured vector This process uses keyword mapping rules, engineering semantic library and context constraints to perform nested parsing. For example, the text "diameter 12mm, two layers staggered arrangement" in the drawing description will be parsed as Rebar dia =12,Layout=2, and then encode into a vector.

[0175] Construction of semantic encoding module for supervised behavior description

[0176] For E ij Text description in D ij , use the pre-trained supervised semantic recognition model to encode and obtain the supervised semantic vector The model is implemented by a Transformer structure, including:

[0177] Input encoding layer (segmenting Chinese professional terms into tokens);

[0178] Multi-head attention encoding layer (used to model dependencies between professional terms);

[0179] The output is the length and G i Aligned vectors

[0180] For example, “the binding has been completed on site, the main reinforcement is 12mm and the spacing is 205mm” will be mapped into a semantic feature vector with the same dimension as the drawing standard.

[0181] Consistency discrimination model design

[0182] Constructing a consistency checking model Its goal is to minimize the difference between the supervised semantics and the drawing standard. The loss function of the model is designed as follows:

[0183]

[0184] in: Represents the Euclidean distance between the drawing specification and the supervision record in the semantic space, measuring the consistency difference; It is a regularization term that encourages the model to avoid overfitting to redundant fields; β is the weight of the regularization term, which is usually set to 0.1-0.2.

[0185] This regularization term can set penalties based on the frequently appearing but non-critical fields in the construction supervision text (such as "site clearing completed"), thereby focusing on the core fields that are strongly related to the drawing content.

[0186] After the model training is completed, two key structural results are output:

[0187] Score the consistency between each node supervision behavior and the drawing standard

[0188] Each construction node R i The total consistency confidence That is, the node verification pass rate.

[0189] These outputs will be passed as input to the next step, “intelligent evaluation and decision execution,” to assist in determining whether supervision is sufficient and whether the supervision task needs to be rescheduled.

[0190] S6. Generate an acceptance rectification strategy based on the output of the supervision consistency model, and convert the acceptance rectification strategy into a specific scheduling execution item;

[0191] This step, the final execution phase of the entire intelligent supervision management process, aims to automatically generate acceptance strategies and corrective actions based on the output of the supervision consistency model constructed in the previous phase, and then convert these strategies into specific scheduling and execution items. By integrating supervision consistency scores, construction node risk levels, and construction duration factors, an executable supervision and correction ranking mechanism is established. The strategy generation module then outputs structured and deployable action recommendations.

[0192] The input is the output of step S5: the consistency score of each supervision behavior and the average confidence score of each node

[0193] At the same time, continue to use the node risk level r obtained in step 1 i and node schedule deadline

[0194] First, the system calculates the number of construction nodes R i Construct an acceptance priority score π i , which is calculated as follows:

[0195]

[0196] Among them, Π i Indicates the acceptance priority score; represents the node supervision consistency score; r i Indicates the node construction risk level; Indicates the node's specified completion time; t now Indicates the current system time; β1, β2, and β3 indicate the preset scheduling policy weights (adjustable, such as 0.5 / 0.3 / 0.2).

[0197] This priority score is used to sort nodes and decide whether each node should undergo automatic acceptance or trigger correction suggestions.

[0198] Two threshold acceptance thresholds θ are set accept and rectification threshold θ fix ;

[0199] like And Π i If the value is less than the preset threshold, an action suggestion that passes the acceptance test will be generated directly;

[0200] like And Π i >θ fix , then generate rectification suggestions and assign rectification execution personnel;

[0201] Otherwise, the node status remains pending. Corrective suggestions and acceptance actions are generated by the strategy generation module, which is built based on the template filling method, such as:

[0202] Acceptance action: "Node R i Recommend immediate acceptance by Supervisor M k implement";

[0203] Corrective action: "Node R i There are supervision defects: the steel bar parameters are inconsistent, it is recommended to ij +Retest within 2 days".

[0204] The output form of the final generated action item is a data structure that can be directly called by the scheduling system, supporting interface deployment and automatic distribution.

[0205] The final output is:

[0206] Acceptance strategy set Each Y i Indicate whether the node has been accepted, who will accept it, and when it will be accepted;

[0207] Corrective Action List Each A i Includes rectification type, responsible person, time plan, operation template, etc.

[0208] See also Figure 2 As shown, in a second aspect of the present invention, there is also provided an intelligent project management system for engineering supervision, comprising:

[0209] The data processing module is used to parse the 2D construction drawings, 3D structural models and design description documents in traditional engineering projects into a set of risk supervision units;

[0210] The risk assessment module is used to evaluate the potential risk change trend of each risk node based on the extracted risk supervision unit set data, and obtain a dynamic risk score set by predicting the risk intensity value of each risk node;

[0211] The risk matching scheduling module is used to calculate the matching degree between each risk node and each supervisor based on the risk intensity score of each risk node, and dynamically adjust the supervisor scheduling plan based on the matching degree score between each supervisor and the node;

[0212] An abnormal data marking and data storage module is configured to structure archive the supervision scheduling plan and the supervision behavior record in the actual construction process, and the system will automatically mark any abnormal supervision behavior and data in the actual construction process;

[0213] A supervision consistency module is configured to build a supervision consistency model, and check the consistency between the supervision behavior, the construction record and the drawing standard;

[0214] An acceptance rectification module is configured to generate an acceptance rectification strategy based on the output result of the supervision consistency model, and convert the acceptance rectification strategy into specific dispatch execution items.

[0215] The above embodiments only describe the preferred embodiments of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by ordinary engineering technicians in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. An intelligent project management method for engineering supervision, characterized in that: The following steps are involved: S1. Analyze the 2D construction drawings, 3D structural models, and design documentation in traditional engineering projects into a set of risk supervision units with engineering structure, construction intent, timing constraints, and supervision priorities. S2. Based on the extracted risk supervision unit set data, evaluate the potential risk change trend of each risk node and obtain a dynamic risk score set by calculating the risk intensity value of each risk node; S3. Based on the risk intensity score of each risk node, calculate the matching degree between each risk node and each supervisor, and dynamically adjust the supervisor scheduling plan based on the matching degree score between each supervisor and the node; S4. The supervisor schedule and the supervisor behavior records during the actual construction process are structured and archived. During the actual construction process, the system will automatically mark any abnormal supervisory behavior and data; S5. Build a supervision consistency model to verify the consistency between supervision behavior, construction records, and drawing standards, and output the consistency score of each node's supervision behavior and drawing standards and the total consistency confidence of each construction node; S6. Generate an acceptance rectification strategy based on the output results of the supervision consistency model, and convert the acceptance rectification strategy into a specific scheduling execution item.

2. The intelligent project management method for engineering supervision according to claim 1, characterized in that: In said S1, in said risk supervision unit set, each risk supervision unit includes component type and number, process description text, structural connectivity and complexity score, priority score and belonging process and planned time.

3. The intelligent project management method for engineering supervision according to claim 1, characterized in that: In S2, the risk intensity value of each risk node is calculated using the following formula: in, Represents the risk node R i The dynamic risk score of , α represents the weighting coefficient, Represents node R i The predicted value of the disturbance at the future time t+Δt, ρ i For node R i risk priority score.

4. The intelligent project management method for engineering supervision according to claim 1, characterized in that: In S3, the matching degree between each risk node and each supervisor is calculated to indicate whether the supervisor is suitable for supervising the node. The higher the matching degree score, the more suitable the supervisor is for the node. The supervisory scheduling plan prioritizes high-risk nodes for supervision, and the constraints include: Each node must have at least one supervisor to supervise; The working hours of each supervisor cannot exceed their maximum available time; Each node must complete supervision within the specified time.

5. The intelligent project management method for engineering supervision according to claim 1, characterized in that: In S4, whenever the supervisor arrives at a construction node and starts supervision, a supervision behavior record is automatically generated, which includes the following information: Action type, including inspection, acceptance, and corrective action recommendations, to indicate the nature of the action; Supervisor number, indicating the person who performs the supervisory task; Construction node number, indicating the construction node corresponding to this supervision behavior; Timestamp, indicating the start and end time of the recorded supervision behavior; Behavior content refers to the specific description of supervisory behavior, including acceptance reports and inspections for compliance with construction specifications.

6. The intelligent project management method for engineering supervision according to claim 1, characterized in that: In S5, the loss function of the supervised consistency model is as follows: in, represents the loss function, G represents the Euclidean distance between the drawing specification and the supervision record in the semantic space, which is used to measure the consistency difference; i Represents a structured vector; Represents the supervised semantic vector; represents the regularization term, which is used to encourage the model to avoid overfitting to redundant fields; β is the weight of the regularization term.

7. The intelligent project management method for engineering supervision according to claim 1, characterized in that: In S6, an acceptance priority score is constructed for each construction node, and the calculation method is as follows: Among them, Π i Indicates the acceptance priority score; represents the node supervision consistency score; r i Indicates the node construction risk level; Indicates the node's specified completion time; t now Indicates the current system time; β1, β2, β3 indicate the preset scheduling strategy weights; The priority score is used to sort nodes and decide whether each node should be automatically accepted or trigger rectification suggestions. Two threshold values ​​are set: acceptance threshold θ accept and rectification threshold θ fix ; like And Π i If the value is less than the preset threshold, an action suggestion that passes the acceptance test will be generated directly; like And Π i >θ fix , then generate rectification suggestions and assign rectification execution personnel; Otherwise, the node status remains pending.

8. An intelligent project management system for engineering supervision, characterized in that: Includes the following connected in sequence: The data processing module is used to parse the 2D construction drawings, 3D structural models and design description documents in traditional engineering projects into a set of risk supervision units; The risk assessment module is used to evaluate the potential risk change trend of each risk node based on the extracted risk supervision unit set data, and obtain a dynamic risk score set by predicting the risk intensity value of each risk node; The risk matching scheduling module is used to calculate the matching degree between each risk node and each supervisor based on the risk intensity score of each risk node, and dynamically adjust the supervisor scheduling plan based on the matching degree score between each supervisor and the node; The abnormal data marking and data storage module is used to structure and archive the supervisory schedule and supervisory behavior records during the actual construction process. During the actual construction process, the system will automatically mark any abnormal supervisory behavior and data; Supervision consistency module, used to build a supervision consistency model to verify the consistency between supervision behavior, construction records and drawing standards; The acceptance and rectification module is used to generate an acceptance and rectification strategy based on the output results of the supervision consistency model, and convert the acceptance and rectification strategy into specific scheduling execution items.

Images