Building industry progress intelligent monitoring method and system based on multi-modal model

Abstract

The invention discloses a building industry progress intelligent monitoring method and system based on a multi-modal model, belongs to the technical field of building construction management, and solves the technical problem of low manual integration and analysis efficiency. The method comprises the following steps: in a planning stage, generating a progress plan through an artificial intelligence algorithm, and determining a project progress plan by personnel adjustment and improvement of a construction unit; in the execution stage, a construction unit starts construction according to a project progress plan, and multi-terminal data acquisition is carried out; in the data processing and integration stage, multi-terminal data is integrated to a unified cloud data platform, and a real scenic spot cloud model is constructed; in the progress recognition and analysis stage, contrastive analysis is conducted on the progress and the BIM, and deviation alarming and deviation evaluation are conducted; in the prediction and early warning stage, the progress completion rate and the risk probability are predicted through a hybrid model; in the monitoring and adjusting stage, the construction process is continuously monitored, an adjusting scheme is automatically generated, and after adjustment, the system enters the execution stage again to form a closed loop. The problems of information splitting and early warning lagging are effectively solved through the multi-modal model.

Description

Technical Field

[0001] This invention relates to the field of construction management technology, and more specifically, to a method and system for intelligent monitoring of construction progress based on a multimodal model. Background Technology

[0002] In the current development of the construction industry, the scale and complexity of construction projects are increasing daily, making accurate monitoring and effective management of project progress a key factor in ensuring successful project delivery. However, existing construction progress monitoring technologies have many problems that urgently need to be addressed.

[0003] First, information fragmentation and inefficient collaboration are prominent issues. In the traditional construction project management model, the construction unit, the construction contractor, and the supervision unit often use their own independent information systems to record and manage project-related information.

[0004] Construction companies typically use project management platforms to control overall project progress, budget, and coordinate relationships among various parties. However, these platforms have relatively low data entry and update frequencies, focusing more on macro-level control. Construction companies use specialized construction management software to record information such as construction progress, manpower allocation, and material distribution. However, this information is limited to the construction phase and lacks effective integration with other companies' information systems. Supervision companies record project quality and safety information through supervision logs and quality inspection systems, but this information is also difficult to share in real time with construction and construction companies.

[0005] For example, when the construction unit adjusts the project investment plan, relevant information cannot be transmitted to the construction and supervision units in a timely and accurate manner. This can lead to situations where the construction unit adjusts the construction schedule due to funding issues, while the supervision unit remains unaware and unable to effectively monitor the project. Similarly, when the construction unit encounters construction difficulties or schedule changes, it cannot promptly synchronize detailed information with the construction and supervision units, making it difficult for the construction unit to grasp the actual progress of the project, and preventing the supervision unit from providing timely professional guidance. This information gap makes it difficult for all parties to obtain comprehensive and accurate project information during project progress monitoring, significantly hindering collaborative work.

[0006] Secondly, there is a lag in progress tracking and early warning. Currently, progress tracking in many construction projects relies mainly on periodic manual reports and on-site inspections. This problem exists for the construction unit, the contractor, and the supervision unit. Construction workers need to report the construction progress to their superiors only every few days or even weeks. During this period, if problems affecting progress occur on-site, such as material shortages or equipment malfunctions, the construction unit and the supervision unit are unlikely to be aware of them in a timely manner.

[0007] The construction unit relies on reports from the construction and supervision units to understand the progress, which is subject to information delays. Although the supervision unit conducts on-site inspections, limitations in the inspection cycle and scope prevent them from identifying all progress risks in real time. Moreover, most existing early warning mechanisms are based on simple threshold settings and lack comprehensive consideration of complex factors. For example, they judge whether the progress is delayed solely based on preset schedule milestones, without taking into account the impact of uncertainties such as weather changes and personnel movement. As a result, in actual projects, when progress risks actually occur, early warnings are often delayed, making it impossible to take timely and effective countermeasures, leading to adverse consequences such as project delays and increased costs.

[0008] Finally, there is a lack of intelligent tools. While intelligent methods have been widely applied across various fields thanks to the rapid development of information technology, their level of intelligence remains low in the construction industry's progress monitoring. Construction companies, contractors, and supervisors still largely rely on traditional tables and charts when processing project progress data.

[0009] When analyzing overall project progress trends, construction companies face a large volume of scattered data from different units, making manual integration and analysis inefficient and hindering the identification of potential problems. Similarly, construction companies lack intelligent tools for recording and analyzing construction progress data, making it impossible to deeply mine the massive amounts of data and identify key factors affecting progress. Supervision companies, lacking intelligent tools, can only rely on experience for judgment in quality and progress correlation analysis, making accurate assessment difficult. Furthermore, the lack of intelligent predictive capabilities prevents accurate prediction of future progress trends based on historical data and real-time information, hindering project managers from developing appropriate contingency plans in advance.

[0010] In summary, existing construction project progress monitoring technologies suffer from serious shortcomings in information integration, progress tracking and early warning, and intelligent applications, failing to meet the demands of efficient management in modern construction projects. Therefore, developing an advanced construction project progress monitoring technology capable of addressing these issues is of significant practical importance. Summary of the Invention

[0011] The technical problem to be solved by the present invention is to address the above-mentioned shortcomings of the prior art. The purpose of the present invention is to provide a method for intelligent monitoring of construction progress based on a multimodal model.

[0012] The second objective of this invention is to provide an intelligent monitoring system for construction progress based on a multimodal model.

[0013] To achieve the first objective mentioned above, this invention provides a method for intelligent monitoring of construction progress based on a multimodal model, comprising: Planning phase: Step 1.1: Develop a project schedule. Using the intelligent schedule development module provided by the system, based on artificial intelligence algorithms, combined with historical data of similar projects, the characteristics of the current project, and resource constraints, a preliminary draft schedule is generated to predict the reasonable construction period for each construction stage. The construction unit personnel adjust and improve the draft schedule based on it, and finally determine the project schedule and upload it to the cloud platform. The system outputs the finalized project schedule. Execution phase: Step 2.1: Construction execution. The construction unit begins construction according to the project schedule, and the system starts real-time monitoring. Step 2.2: Multi-terminal data acquisition, Construction unit side: Investment plans and decision-making information are entered through project management software, and the data is uploaded to the cloud platform in real time; On the construction unit side: Deploy high-definition cameras and laser scanners to collect construction videos and point cloud data; at the same time, collect personnel data through attendance gates and smart safety helmets, and upload the data to the cloud platform after preprocessing. Supervision unit side: Use the supervision APP to enter quality inspection and safety record data, and upload the data to the cloud platform; testing instruments automatically upload data to the cloud platform; Data processing and integration phase: Step 3.1: Data preprocessing and integration. The collected data is integrated into a unified cloud data platform through standardized formats and interface technologies to perform data cleaning and format unification to ensure data availability. Step 3.2: Site environment identification and construction. Using computer vision technology to process image and point cloud data, and through the TSDF 3D reconstruction algorithm, a real-world point cloud model of the construction site is generated, and a real-time 3D model of the construction site is output. Progress identification and analysis phase: Step 4.1: Comparative analysis with the BIM model. Align and match the constructed real-world cloud model with the BIM model in terms of coordinates. Calculate the comprehensive similarity score Sim_score based on geometric overlap ratio IoU, structural similarity SSIM, and personnel configuration matching degree People_match. If Sim_score < threshold, trigger a deviation alarm. Step 4.2: Schedule Deviation Assessment. Based on the comparative analysis results, accurately identify the deviation between the current schedule and the planned schedule, and output the current schedule status and deviation report; Forecasting and Early Warning Phase: Step 5.1: Progress prediction. Using an LSTM+Prophet hybrid model, input real-time data to predict the progress completion rate at a future time (t+1). The model is optimized based on historical data during training, with a target MAE ≤ 3%. Step 5.2: Risk probability calculation and early warning, based on the current deviation. Calculate the risk probability using the prediction deviation ΔS ; if If the threshold is exceeded, the system will automatically issue a tiered warning, output the warning information, and notify the relevant units. Monitoring and Adjustment Phase: Step 6.1: Real-time monitoring. The system continuously monitors the construction process, integrates multimodal data, and combines big data analysis to promptly identify potential problems. Step 6.2: Adjustment plan generation and execution. Based on monitoring results and early warning information, the system automatically generates an adjustment plan. The construction unit and the contractor will make decisions based on the plan and adjust the schedule or execution method accordingly. After the adjustment, the system re-entered the execution phase, continued data collection and monitoring, and formed a closed loop.

[0014] As a further improvement, in step 4.1,

[0015] in, To calculate the overall similarity score, For spatiotemporal matching, For process matching, For personnel matching, , , Let be the weighting coefficient, satisfying .

[0016] Furthermore, The calculation process is as follows: Spatial division: The planned construction area of ​​the BIM model is divided into the smallest calculation unit according to "construction section + floor / part"; Point cloud filtering: Filter out valid real point clouds within the computing unit from the 3D scene model of the construction site; Volume Calculation: Calculate the volume of each BIM planning unit. The volume of the real-world cloud ; Overlap calculation: ,in It is the volume of the overlapping region between the two. It is the volume of the total area covered by both.

[0017] Furthermore, The calculation process is as follows: Calculate the consistency of process sequence: Extract the process dependencies from the BIM plan and construct the process chain. Determining the actual process chain based on the geometric features of real-world point clouds. ; Computational consistency: ; Calculate structural detail consistency: Extract key structural details from the BIM plan to obtain plan parameter i; measure the actual parameter i corresponding to the key structural details from the real-world point cloud; calculate consistency: ; For the number of detailed parameters; final: ; , Let be the weighting coefficient, satisfying .

[0018] Furthermore, The calculation process is as follows: Calculate the job matching rate: ; Calculate the number of people meeting the target: ; Calculate efficiency matching rate: ; final: ; , , Let be the weighting coefficient, satisfying .

[0019] Furthermore, in step 5.1, a hybrid model is constructed using LSTM+Prophet:

[0020] in; This represents the actual progress completion rate at time t+1. =[Schedule Vector, Resource Utilization Rate, Process Duration] The efficiency reduction coefficient due to weather conditions. Supply chain risk level, This is the random error term.

[0021] Further, in step 5.2, a risk prediction model is constructed:

[0022]

[0023] in, = , The planned progress completion rate at time t+1. For risk sensitivity coefficient, , Let be the weighting coefficient, satisfying The risk probability is obtained by developing a risk prediction model.

[0024] To achieve the first objective mentioned above, this invention provides a construction industry progress intelligent monitoring system based on a multimodal model, including a construction unit terminal, a construction contractor terminal, a supervision unit terminal, and a cloud data platform. The cloud data platform connects to the construction unit terminal, the construction contractor terminal, and the supervision unit terminal through a secure and encrypted network. Construction unit personnel input project investment plans and important decision-making information through the construction unit's terminal, while the cloud data platform automatically collects macro progress data from the construction unit's project management platform on a regular basis through the construction unit's terminal. The cloud-based data platform collects construction videos from high-definition cameras, point cloud data from laser scanners, and personnel data from attendance gates and smart safety helmets from the construction unit. The cloud-based data platform collects quality inspection and safety record data, as well as data automatically uploaded by testing instruments, through the supervision unit's end. The cloud-based data platform monitors construction progress using the aforementioned intelligent monitoring method for the construction industry based on a multimodal model.

[0025] Beneficial effects Compared with the prior art, the present invention has the following advantages: 1. Data Acquisition and Integration: By setting up various data acquisition devices and software at the construction and supervision units, comprehensive data on construction progress, quality, safety, and other aspects can be collected. Standardized formats and data interface technologies are then used to integrate this data into a unified platform, which forms the basis for subsequent intelligent monitoring.

[0026] 2. Multimodal Model Recognition: Utilizing data acquired from devices such as cameras and laser scanners, a 3D scene model of the construction site is constructed and compared with the BIM model to accurately identify schedule deviations. Simultaneously, schedule risks are predicted based on big data and machine learning algorithms; this is the core technology for accurate tracking and early warning of progress.

[0027] 3. Closed-loop system construction: Create a closed-loop system of "planning-execution-monitoring-adjustment" to make project progress management an organic whole, with each link being interconnected and mutually influential, so as to achieve dynamic control.

[0028] 4. Unique data acquisition and integration methods: covering the layout of acquisition devices at each end, the types and frequencies of data acquisition, data preprocessing methods, and the specific processes and technologies for integrating data into a unified platform, preventing other entities from using similar data acquisition and integration models without authorization.

[0029] 5. Application of multimodal model recognition technology: including algorithms for identifying and constructing site environment, specific methods for comparative analysis with BIM models, training methods and parameter settings for progress prediction models, etc., protecting the innovative application of this technology in the field of construction progress monitoring, and effectively solving the problems of information fragmentation and delayed early warning.

[0030] 6. The architecture and operational logic of the closed-loop system: including the algorithmic logic of the planning module, the specific indicators and methods for execution monitoring, and the rules for generating adjustment plans, to ensure that this innovative project schedule management closed-loop system cannot be imitated. Attached Figure Description

[0031] Figure 1 This is a diagram of the architecture of the present invention; Figure 2 This is a flowchart of the construction site environment identification and construction process of the present invention; Figure 3 A schematic diagram for comparison and analysis with the BIM model; Figure 4 This is a flowchart of a closed-loop system of "planning-execution-monitoring-adjustment". Detailed Implementation

[0032] The present invention will be further described below with reference to specific embodiments shown in the accompanying drawings.

[0033] See Figures 1-4 A method for intelligent monitoring of construction progress based on a multimodal model includes the following steps in each stage: Planning phase: Step 1.1: Develop a project schedule. Using the intelligent scheduling module provided by the system, based on artificial intelligence algorithms and combined with historical data from similar projects, the characteristics of the current project, and resource constraints, a preliminary draft schedule is generated. For example, based on factors such as the project's building area, building type, and local climate conditions, the reasonable construction period for each stage is predicted. The construction unit personnel adjust and improve the draft schedule, ultimately determining the project schedule and uploading it to the cloud platform. The system outputs the finalized project schedule (as a baseline).

[0034] Execution phase (construction commencement and data collection): Step 2.1: Construction Execution. The construction unit begins construction according to the project schedule, and the system initiates real-time monitoring. During the monitoring process, data collected from the construction unit is analyzed to optimize the allocation of construction resources. For example, based on the construction progress and equipment usage, the system rationally schedules the deployment of construction equipment to avoid idle or overused equipment. Simultaneously, the system evaluates the work efficiency of construction personnel. If a construction team is found to have low efficiency, the system promptly alerts the construction unit's management to take measures, such as providing skills training or adjusting personnel arrangements.

[0035] Step 2.2: Multi-terminal data acquisition, On the construction unit side: Investment plans and decision-making information (such as budget adjustments) are entered through project management software, and the data is uploaded to the cloud platform in real time.

[0036] On the construction unit side: Deploy high-definition cameras and laser scanners to collect construction videos and point cloud data; at the same time, collect personnel data through attendance gates and smart safety helmets, and upload the data to the cloud platform after preprocessing.

[0037] On the supervision unit's end: Use the supervision APP to enter quality inspection and safety record data, and upload the data to the cloud platform; the testing instruments automatically upload data to the cloud platform.

[0038] Key data specifications include: image acquisition (hardware configuration, frequency, lighting requirements), multi-view stereo matching (generating depth maps), and TSDF 3D reconstruction (generating standardized point cloud models).

[0039] Data processing and integration phase: Step 3.1: Data preprocessing and integration. The collected data is integrated into a unified cloud data platform through standardized formats and interface technologies to perform data cleaning and format unification to ensure data availability.

[0040] Step 3.2: Site environment identification and construction. Use computer vision technology (such as object detection algorithm) to process image and point cloud data, and generate a real-time point cloud model of the construction site (i.e., a 3D scene model) through the TSDF 3D reconstruction algorithm, and output the real-time 3D model of the construction site.

[0041] Progress identification and analysis phase: Step 4.1: Comparative analysis with the BIM model. Align and match the constructed real-world cloud model with the BIM model (digital blueprint) in terms of coordinates. Calculate the comprehensive similarity score Sim_score based on geometric overlap ratio IoU, structural similarity SSIM, and personnel configuration matching degree People_match. If Sim_score < threshold, for example, Sim_score < 0.8, trigger a deviation alarm.

[0042] Step 4.2: Schedule Deviation Assessment. Based on the comparative analysis results, accurately identify the deviation between the current schedule and the planned schedule (such as position and degree), and output the current schedule status and deviation report.

[0043] The system also performs comprehensive analysis of various progress-related data to promptly identify potential problems. For example, by analyzing material supply data and construction progress data, it can determine whether there is a risk of delays due to untimely material supply. Simultaneously, it monitors safety and quality data at the construction site, and if any safety hazards or quality issues are discovered, it promptly notifies relevant departments for handling, ensuring that the construction progress is not affected.

[0044] Forecasting and Early Warning Phase: Step 5.1: Progress prediction, using an LSTM+Prophet hybrid model, inputting real-time data (such as Sim_score, time series data). , The efficiency reduction coefficient due to weather conditions. Based on the supply chain risk level, predict the progress completion rate (such as cumulative completion) at a future time (t+1); optimize the model based on historical data during the training process, with a target MAE ≤ 3%.

[0045] Step 5.2: Risk probability calculation and early warning, based on the current deviation. Calculate the risk probability using the prediction deviation ΔS .

[0046] if If the threshold is exceeded, the system will automatically issue a graded warning (including risk type, degree, and response suggestions), output the warning information, and notify the relevant units (construction unit, construction unit, and supervision unit).

[0047] Monitoring and Adjustment Phase: Step 6.1: Real-time monitoring. The system continuously monitors the construction process, integrates multimodal data (such as progress, quality, and safety), and combines big data analysis to promptly identify potential problems (such as material supply delays).

[0048] Step 6.2: Adjustment Plan Generation and Execution. Based on monitoring results and early warning information, the system automatically generates adjustment plans (such as adjusting the construction sequence or increasing resources). The adjustment plans consider various factors, such as remaining construction period, resource availability, and construction process requirements. For example, when the system predicts a delay in the progress of a certain part of the project, it generates plans to adjust the construction sequence, increase construction personnel or equipment, etc., based on the current resource situation and subsequent construction tasks. The construction unit and the contractor can make decisions based on the adjustment plans provided by the system and the actual situation, adjusting the project schedule to achieve dynamic control over the project progress.

[0049] The construction unit and the contractor will make decisions based on the plan and adjust the schedule or execution method accordingly.

[0050] After adjustment, the system re-enters the execution phase (step 2.1) to continue data collection and monitoring, forming a closed loop. This enables a closed-loop system of "planning-execution-monitoring-adjustment," making project schedule management an organic whole where each link is interconnected and mutually influential, achieving dynamic control.

[0051] In step 2.2, the data acquisition specifications are as follows: (1) Image acquisition requirements Hardware configuration: It adopts an industrial-grade high-definition camera (pixels ≥ 12 million, frame rate ≥ 30fps) and supports multi-view synchronous shooting (at least 3 shooting angles, covering the entire construction area). Data collection frequency: Data is collected according to construction progress nodes (such as before the end of each workday, weekly nodes), and data collection is intensified in core areas (main structure, key processes); Data Acquisition Specifications: Ensure uniform lighting during shooting (avoid backlighting / excessive shadows), maintain a distance of ≤15m between the lens and the construction area, and ensure an overlap rate of ≥60% between adjacent viewpoints to guarantee MVSNet matching accuracy.

[0052] (2) Multi-view stereo matching processing Core functions: Perform feature extraction, matching, and depth estimation on multi-view images to generate high-density depth maps; Technical parameters: Matching window size 3×3~7×7, depth map resolution ≥1024×768, depth estimation error ≤2cm (for targets within 10m); Output results: Single-view depth map, dense point cloud after multi-view consistency verification (preliminary denoising, removal of isolated points).

[0053] (3) TSDF 3D Reconstruction Optimization Core logic: Based on the truncated signed distance function, multi-view depth maps are fused using volumetric fusion to generate a continuous real-world point-cloud model; Key parameters: Voxel resolution ≤ 2cm, truncation distance ≤ 5cm, point cloud density ≥ 500pt / m³ (to ensure detail reproduction and meet the requirements for geometric overlap calculation). Post-processing steps: Through statistical filtering (removing outliers) and downsampling (optimizing computational efficiency while maintaining density), a standardized point cloud model is finally output.

[0054] In step 4.1,

[0055] in, To calculate the overall similarity score, For spatiotemporal matching, For process matching, For personnel matching, , , Let be the weighting coefficient, satisfying Specifically, the proportions can be set according to the importance of the business to adjust the degree of influence of the three-dimensional indicators (e.g., prioritizing geometric precision). (Let's set it to be larger). This is the final evaluation result, and its value typically ranges from 0 to 1 (the closer to 1, the higher the match). When When the value is less than the threshold, it indicates that the deviation between the two objects exceeds the acceptable range, triggering an alarm. For example, if the threshold is 0.8, and Sim_score < 0.8, it means that the deviation between the two objects exceeds the acceptable range, triggering an alarm (such as manual review, process suspension, etc.).

[0056] The advantages of the model are as follows: Unified variable dimensions: IoU (spatiotemporal matching), SSIM (process matching), and People_match (personnel matching) are all matching indicators of "actual vs. planned", with consistent dimensions (0-1), and the weighted summation logic holds true; Based on actual site conditions: Over 80% of construction progress deviations are related to "personnel allocation" (labor shortages, mismatch of job types, low efficiency). The formula can cover the "human" factor and is more practical. More precise alarm triggering: For example, a project has an IoU of 0.8 (compliance with completion targets) and an SSIM of 0.9 (compliance with work procedures), but People_match = 0.5 (3 core workers are missing). The overall Sim_score is 0.5 × 0.8 + 0.3 × 0.9 + 0.2 × 0.5 = 0.77 < 0.8, triggering an alarm and providing early warning of the risk of subsequent progress delays (the original formula would miss alarms due to IoU and SSIM meeting targets).

[0057] The calculation process is as follows: Spatial division: Divide the planned construction area of ​​the BIM model into the smallest calculation unit according to "construction section + floor / part" (e.g., "3-5 axis × 2 floors"). Point cloud filtering: Filter out valid real-world point clouds within the computing unit from the 3D scene model of the construction site (removing irrelevant points such as ground and temporary facilities); Volume Calculation: Calculate the volume of each BIM planning unit. The volume of the real-world cloud (Based on convex hull algorithm or voxelization method); Overlap calculation: ,in It is the volume of the overlapping region between the two. It is the volume of the total area covered by both.

[0058] For example, the planned unit volume =100m³, actual point cloud volume =85m³, overlapping volume =80m³, then IoU=80 / (100+85-80)=80 / 105≈0.76.

[0059] The calculation process is as follows: Calculate the consistency of process sequence: Extract the process dependencies from the BIM plan and construct the process chain. (e.g., "formwork erection → rebar tying → pouring"); The actual process chain is determined based on the geometric features of the actual site cloud. (e.g., the distribution of reinforcement point clouds, the smoothness of the concrete surface); Calculation consistency: If the plan involves 3 steps, and the actual process is perfectly matched... =1.0; reverse 1 step and... =0.67 (two steps are corresponding).

[0060] Calculate structural detail consistency: Extract key structural details from the BIM plan (such as rebar spacing, pipeline slope, and component cross-sectional dimensions) to obtain the planned parameter i; measure the actual parameter i of the corresponding key structural details from the actual point cloud (such as measuring rebar spacing through point cloud slices); calculate consistency: ; This refers to the number of detailed parameters.

[0061] final: ; , Let be the weighting coefficient, satisfying In this embodiment, , .

[0062] For example, the process sequence matches perfectly ( =1.0), the deviations of the three detailed parameters are 3%, 5%, and 8% respectively. =(1-0.03)×(1-0.05)×(1-0.08)=0.97×0.95×0.92≈0.85), then SSIM=0.6×1.0 +0.4×0.85=0.6+0.34=0.94.

[0063] Quantifying the alignment between actual staffing (job types, number of people, efficiency) and the "resource allocation plan" in the BIM plan is the core reflection of "sustainability of schedule execution". The calculation process is as follows: Calculate the job matching rate: Correction rules: No redundant jobs → correction coefficient is 1.0; There are unnecessary redundant jobs → correction coefficient is 0.8; There is a shortage of key jobs (such as steel reinforcement workers) → correction coefficient is 0.5 (if the shortage of steel reinforcement workers is greater than 50%, then the correction coefficient is 0.0).

[0064] Calculate the number of people meeting the target: The main job qualification coefficient is as follows: 1.0 for job vacancies (number of vacant positions) ≤ 10%; 0.9 for job vacancies ≤ 20%; and 0.7 for job vacancies > 20%.

[0065] Calculate efficiency matching rate: Efficiency units should be consistent with the plan, such as tons / person / day or square meters / person / day.

[0066] final: ; , , Let be the weighting coefficient, satisfying In this embodiment, , , .

[0067] For example, the plan is to have 2 types of workers (5 steelworkers and 3 general laborers), but the actual number of workers on duty is 2 matching types (4 steelworkers and 3 general laborers), with no redundancy. =2 / 2×1.0=1.0); Total staffing target achievement rate = 7 / 8=0.875, 20% shortage in main job categories (coefficient 0.9) → =0.875×0.9≈0.788; The planned average daily rebar tying volume for steelworkers was 5 tons, but the actual volume was 4.5 tons. =4.5 / 5=0.9; final =0.4×1.0 + 0.3×0.788 + 0.3×0.9≈0.91.

[0068] In step 5.1, a hybrid model is constructed using LSTM+Prophet:

[0069] in; This represents the actual progress completion rate at time t+1. =[Schedule Vector, Resource Utilization Rate, Process Duration] The efficiency reduction coefficient due to weather conditions. Supply chain risk level, This represents the random error term. It combines the time-dependent capture capabilities of LSTM with the nonlinear fitting capabilities of Prophet for risk factors, thereby improving prediction accuracy and precisely predicting the progress at time t+1.

[0070] Capture temporal dependencies and output basic prediction values; Input layer: =[Schedule Vector, Resource Utilization, Process Duration] (Constructing a time series window from historical 7-30 days of data); Adjustment items: (Weather-induced efficiency reduction coefficient: 0.6-1.0) (Supply chain risk level: 1-5, quantified as 0.2-1.0); Network structure: 2 layers of LSTM (64 / 32 hidden units) + 1 fully connected layer, with ReLU activation function, outputting basic prediction values.

[0071] Focusing on the nonlinear effects of risk factors, output correction values; Core function: To compensate for the insufficient fitting of LSTM to "sudden risks" (such as rainstorms and material supply disruptions). The trend term, seasonal term, and holiday term are decomposed using an additive model, with a focus on fitting the data. and The impact of sudden changes (such as the reduction in efficiency when precipitation is ≥50mm / d). Output: Schedule deviation correction value caused by risk factors (e.g., "Heavy rain caused a 2% delay in schedule, after correction, \hat{Y}_{t+1} is reduced by 2%"). ε (random error term): covers sudden factors not captured by the model (such as temporary leave of personnel or equipment failure), with a value range of ±1% (based on historical data statistics).

[0072] In step 5.2, a risk prediction model is constructed:

[0073]

[0074] in, = (e.g., 0.75) This represents the planned progress completion rate at time t+1 (e.g., 8%). The default value for the risk sensitivity coefficient is 2.5 (which can be adjusted according to project type). The larger, The more significant the impact of S on P_risk (e.g., in precision engineering), the greater the impact of S on P_risk. =3.0, Ordinary Engineering =2.0), , Let be the weighting coefficient, satisfying In this embodiment, , The risk probability is obtained by developing a risk prediction model.

[0075] For example, it is known that: =0.75 (current deviation 0.05), =8%, =5% (prediction bias 3%) =2.5; calculate : 0.6×(0.8-0.75) + 0.4×(8%-5%)=0.6×0.05 + 0.4×0.03=0.03+0.012=0.042; calculate : 1 / (1+e^(-2.5×0.042))=1 / (1+e^(-0.105))≈1 / (1+0.9)=0.526 (Risk probability 52.6%, no warning required).

[0076] A construction industry progress intelligent monitoring system based on a multimodal model includes a construction unit terminal, a construction contractor terminal, a supervision unit terminal, and a cloud data platform. The cloud data platform connects to the construction unit terminal, the construction contractor terminal, and the supervision unit terminal through a secure and encrypted network.

[0077] The construction unit is equipped with dedicated project management software client, which connects to the cloud data platform via a secure and encrypted network. Construction unit personnel input project investment plans, key decision-making information, and other data into the software. For example, when the construction unit decides to increase or decrease the budget for a sub-project, the relevant data is uploaded to the cloud platform in real time. Simultaneously, the system automatically collects macro-level progress data from the construction unit's project management platform periodically; that is, the cloud data platform automatically collects macro-level progress data from the construction unit's project management platform through the construction unit's client, such as the completion status of overall project milestones.

[0078] The cloud-based data platform collects construction videos from high-definition cameras, point cloud data from laser scanners, and personnel data from attendance gates and smart safety helmets from the construction unit. Various types of data acquisition devices are deployed at the construction site, with high-definition cameras installed in various construction areas to capture video footage of the construction process, recording construction procedures and the operation of personnel and equipment. Attendance gates are installed at construction entrances and exits, and personnel are required to wear smart safety helmets. This data from different devices and personnel is pre-processed by a data preprocessing module and transmitted to the cloud-based data platform via wireless or wired networks according to a unified data format.

[0079] The cloud-based data platform collects quality inspection and safety record data from the supervision unit, as well as data automatically uploaded by testing instruments. Supervision personnel use a dedicated supervision app to promptly input information such as quality inspection data and safety hazard records during routine inspections. This includes, for example, the concrete strength test results of building structures and the inspection status of safety protection measures at the construction site. Simultaneously, the supervision unit's quality testing instruments connect to the app via Bluetooth or a network interface, automatically uploading test data. This data is also aggregated on the cloud-based data platform, achieving data integration across all terminals.

[0080] The cloud-based data platform monitors construction progress using the aforementioned intelligent monitoring method for the construction industry based on a multimodal model. This constitutes a closed-loop system of "planning-execution-monitoring-adjustment".

[0081] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention, and these will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. A method for intelligent monitoring of construction progress based on a multimodal model, characterized in that, include: Planning phase: Step 1.1: Develop a project schedule. Using the intelligent schedule development module provided by the system, based on artificial intelligence algorithms, combined with historical data of similar projects, the characteristics of the current project, and resource constraints, a preliminary draft schedule is generated to predict the reasonable construction period for each construction stage. The construction unit personnel adjust and improve the draft schedule based on it, and finally determine the project schedule and upload it to the cloud platform. The system outputs the finalized project schedule. Execution phase: Step 2.1: Construction execution. The construction unit begins construction according to the project schedule, and the system starts real-time monitoring. Step 2.2: Multi-terminal data acquisition, Construction unit side: Investment plans and decision-making information are entered through project management software, and the data is uploaded to the cloud platform in real time; On the construction unit side: Deploy high-definition cameras and laser scanners to collect construction videos and point cloud data; at the same time, collect personnel data through attendance gates and smart safety helmets, and upload the data to the cloud platform after preprocessing. Supervision unit side: Use the supervision APP to enter quality inspection and safety record data, and upload the data to the cloud platform; The testing instrument automatically uploads data to the cloud platform; Data processing and integration phase: Step 3.1: Data preprocessing and integration. The collected data is integrated into a unified cloud data platform through standardized formats and interface technologies to perform data cleaning and format unification to ensure data availability. Step 3.2: Site environment identification and construction. Using computer vision technology to process image and point cloud data, and through the TSDF 3D reconstruction algorithm, a real-world point cloud model of the construction site is generated, and a real-time 3D model of the construction site is output. Progress identification and analysis phase: Step 4.1: Comparative analysis with the BIM model. Align and match the constructed real-world cloud model with the BIM model in terms of coordinates. Calculate the comprehensive similarity score Sim_score based on geometric overlap ratio IoU, structural similarity SSIM, and personnel configuration matching degree People_match. If Sim_score < threshold, trigger a deviation alarm. Step 4.2: Schedule Deviation Assessment. Based on the comparative analysis results, accurately identify the deviation between the current schedule and the planned schedule, and output the current schedule status and deviation report; Forecasting and Early Warning Phase: Step 5.1: Progress prediction. Using an LSTM+Prophet hybrid model, input real-time data to predict the progress completion rate at a future time (t+1). The model is optimized based on historical data during training, with a target MAE ≤ 3%. Step 5.2: Risk probability calculation and early warning, based on the current deviation. Calculate the risk probability using the prediction deviation ΔS ; if If the threshold is exceeded, the system will automatically issue a tiered warning, output the warning information, and notify the relevant units. Monitoring and Adjustment Phase: Step 6.1: Real-time monitoring. The system continuously monitors the construction process, integrates multimodal data, and combines big data analysis to promptly identify potential problems. Step 6.2: Adjustment plan generation and execution. Based on monitoring results and early warning information, the system automatically generates an adjustment plan. The construction unit and the contractor will make decisions based on the plan and adjust the schedule or execution method accordingly. After the adjustment, the system re-entered the execution phase, continued data collection and monitoring, and formed a closed loop.

2. The intelligent monitoring method for construction industry progress based on a multimodal model according to claim 1, characterized in that, In step 4.1, in, To calculate the overall similarity score, For spatiotemporal matching, For process matching, For personnel matching, , , Let be the weighting coefficient, satisfying .

3. The intelligent monitoring method for construction industry progress based on a multimodal model according to claim 2, characterized in that, The calculation process is as follows: Spatial division: The planned construction area of ​​the BIM model is divided into the smallest calculation unit according to "construction section + floor / part"; Point cloud filtering: Filter out valid real point clouds within the computing unit from the 3D scene model of the construction site; Volume Calculation: Calculate the volume of each BIM planning unit. The volume of the real-world cloud ; Overlap calculation: ,in It is the volume of the overlapping region between the two. It is the volume of the total area covered by both.

4. The intelligent monitoring method for construction industry progress based on a multimodal model according to claim 2, characterized in that, The calculation process is as follows: Calculate the consistency of process sequence: Extract the process dependencies from the BIM plan and construct the process chain. ; Determining the actual process chain based on the geometric features of real-world point clouds. ; Computational consistency: ; Calculate structural detail consistency: Extract key structural details from the BIM plan to obtain plan parameter i; measure the actual parameter i corresponding to the key structural details from the real-world point cloud; calculate consistency: ; For the number of detailed parameters; final: ; , Let be the weighting coefficient, satisfying .

5. The intelligent monitoring method for construction industry progress based on a multimodal model according to claim 2, characterized in that, The calculation process is as follows: Calculate the job matching rate: ; Calculate the number of people meeting the target: ; Calculate efficiency matching rate: ; final: ; , , Let be the weighting coefficient, satisfying .

6. The intelligent monitoring method for construction industry progress based on a multimodal model according to claim 1, characterized in that, In step 5.1, a hybrid model is constructed using LSTM+Prophet: in; This represents the actual progress completion rate at time t+1. =[Schedule Vector, Resource Utilization Rate, Process Duration] The efficiency reduction coefficient due to weather conditions. Supply chain risk level, This is the random error term.

7. The intelligent monitoring method for construction industry progress based on a multimodal model according to claim 6, characterized in that, In step 5.2, a risk prediction model is constructed: in, = , The planned progress completion rate at time t+1. For risk sensitivity coefficient, , Let be the weighting coefficient, satisfying The risk probability is obtained by developing a risk prediction model.

8. A construction industry progress intelligent monitoring system based on a multimodal model, characterized in that, It includes the construction unit's end, the construction contractor's end, the supervision unit's end, and the cloud data platform. The cloud data platform connects the construction unit's end, the construction contractor's end, and the supervision unit's end through a secure and encrypted network. Construction unit personnel input project investment plans and important decision-making information through the construction unit's terminal, while the cloud data platform automatically collects macro progress data from the construction unit's project management platform on a regular basis through the construction unit's terminal. The cloud-based data platform collects construction videos from high-definition cameras, point cloud data from laser scanners, and personnel data from attendance gates and smart safety helmets from the construction unit. The cloud-based data platform collects quality inspection and safety record data, as well as data automatically uploaded by testing instruments, through the supervision unit's end. The cloud data platform monitors the construction progress using the intelligent monitoring method for construction industry progress based on a multimodal model, as described in any one of claims 1-7.

Images