Intelligent construction resource dynamic scheduling optimization method based on industrial data analysis

Abstract

The application discloses an intelligent construction resource dynamic scheduling optimization method based on industrial data analysis, and aims to solve the problems of material quantity monitoring lag, stoppage for waiting for materials caused by supply-demand mismatch, and low resource utilization in large-scale building indoor construction. The application allocates a unique independent identity code of a target construction area to a construction worker, synchronously collects construction progress and material quantity data through a monitor, completes material supply-demand matching determination through an industrial data analysis engine, adaptively triggers material supplement or cross-region redundant material scheduling, and synchronously realizes construction illegal behavior monitoring. The application realizes material data-driven dynamic scheduling, effectively avoids stoppage for waiting for materials, improves material utilization and construction control efficiency, and is suitable for large-scale indoor construction scenes.

Description

Technical Field

[0001] This application relates to the field of construction resource management methods, and in particular to a method for intelligent construction resource dynamic scheduling and optimization based on industrial data analysis. Background Technology

[0002] New-type industrialized construction and intelligent construction are the core development directions for the transformation and upgrading of my country's construction industry. Interior decoration and electromechanical installation in large-scale residential communities and other buildings are key aspects of schedule, cost, and quality control throughout the entire construction process. Precise and efficient scheduling of construction materials is crucial for ensuring the continuity of interior construction, shortening project timelines, and reducing construction costs. Meanwhile, industrial data analytics, as a core supporting technology for industrial digital transformation, has formed a mature technical system and application paradigm in resource optimization scheduling, precise process control, and closed-loop data management throughout the industrial production field. This provides important technical reference and a foundation for the digital upgrading of resource scheduling in the construction industry.

[0003] Currently, in the field of construction resource scheduling, the conventional approach is a crude management model of centralized procurement and batch distribution by contract section. Material scheduling plans rely heavily on the construction experience of on-site management personnel. Existing digital material scheduling systems mostly rely on the project's pre-set static construction schedule for material demand planning and scheduling. A few optimization solutions involve deploying fixed visual monitoring equipment at the construction site to monitor material inventory in specific centralized storage areas. In addition, existing technologies have solutions that apply industrial data analysis methods to the scheduling of bulk materials and the operation and maintenance management of large construction equipment during the main construction phase, which have achieved digital optimization of construction resource scheduling to a certain extent.

[0004] However, existing technologies still have many insurmountable technical shortcomings for large-scale building indoor construction scenarios: First, indoor construction is characterized by dispersed work areas, small material requirements in individual areas, and significant differences in construction progress between areas. Fixed monitoring equipment is costly to deploy, has many blind spots, and cannot accurately obtain the remaining material quantity in each work area. It can only rely on feedback from construction personnel after the fact regarding material shortages, which can easily lead to work stoppages due to material shortages, and thus trigger a chain reaction of project delays. Second, existing scheduling schemes cannot dynamically calculate the actual material demand in conjunction with the construction progress, resulting in low accuracy in material supply and demand matching. This can easily lead to problems such as material accumulation and waste on site, ineffective and repeated transportation, and low comprehensive utilization rate of material resources. Third, existing schemes cannot achieve integrated integration of material resource scheduling and construction process safety and compliance management. They require the deployment of an additional independent construction safety monitoring system, resulting in high on-site management costs. At the same time, they lack a model self-optimization mechanism adapted to dispersed indoor construction scenarios, and the scheduling accuracy and scenario adaptability are difficult to meet the batch application needs of large-scale projects. Summary of the Invention

[0005] The purpose of this invention is to address the problems of lagging material surplus monitoring, supply and demand mismatch leading to work stoppages due to material shortages, and low resource utilization in the indoor construction of large-scale buildings. This application provides a method for dynamic scheduling and optimization of intelligent construction resources based on industrial data analysis.

[0006] To achieve the above objectives, the intelligent construction resource dynamic scheduling optimization method based on industrial data analysis provided in this application adopts the following technical solution:

[0007] A method for dynamic scheduling and optimization of intelligent construction resources based on industrial data analysis includes:

[0008] Each construction worker involved in indoor construction is assigned a unique identification code. The unique identification code is then bound to the target construction area, preset construction procedures, and standard material requirement list corresponding to that construction worker, forming bound data. This bound data is then stored in the local database of the scheduling controller.

[0009] Each construction worker is equipped with a monitor that is uniquely bound to their own unique identification code. The monitor's field of view is controlled to rotate synchronously with the construction worker's field of view, collect video data streams of the construction scene, and transmit the video stream data carrying the corresponding unique identification code to the scheduling controller.

[0010] The scheduling controller parses and processes the received video data stream, identifies the completed process nodes within the current target construction area based on a pre-trained construction process image recognition model, and calculates the construction progress data.

[0011] When the monitor detects that its field of view is directed toward the fixed material storage area within the target construction area, the remaining quantity calculation process is triggered. Based on the pre-trained material category identification model and remaining quantity calculation model, the remaining quantity data of each category of material in the fixed material storage area is extracted.

[0012] The scheduling controller takes the standard material requirements list and construction progress data as input, calculates the theoretical material requirements required to complete the remaining construction procedures in the current target construction area, matches and verifies the theoretical material requirements with the remaining quantity data of the corresponding material categories, and generates a material supply and demand matching judgment result.

[0013] Based on the material supply and demand matching determination result, if the current remaining material quantity is insufficient to support the completion of the remaining construction processes, the scheduling controller generates a material replenishment scheduling instruction for the corresponding category and quantity, and transfers the required materials to the fixed material storage area of ​​the target construction area through the material transfer unit; if it is identified that there are redundant materials of the same type in the target construction area that exceed the demand of its remaining processes, the scheduling controller generates a cross-regional material scheduling instruction, and transfers the redundant materials to the target construction area with insufficient remaining material quantity.

[0014] Preferably, each construction worker involved in indoor construction is assigned a unique identification code, which is then linked to the target construction area, pre-defined construction procedures, and standard material requirements list for that worker. Specifically, this includes:

[0015] Based on the BIM model of the target construction area, the schedule of the pre-set construction procedures, and the material consumption quota per unit procedure, calculate the standard material usage for a single procedure and the total usage for the entire procedure, and generate a standard material requirement list for the corresponding target construction area.

[0016] The skill level information and safety training record information of the corresponding construction personnel are synchronously bound and stored with their unique identification codes.

[0017] Preferably, the video data stream collected from the construction scene specifically includes:

[0018] The monitor collects head posture data of construction workers, and the monitor is driven to adjust the shooting angle synchronously with the head rotation of the construction workers, so that the image collected by the monitor matches the field of vision of the construction workers.

[0019] The monitor adds a unique identifier to the collected video stream data and transmits the identified video stream data to the scheduling controller via a wireless LAN.

[0020] Preferably, the calculated construction progress data specifically includes:

[0021] A pre-trained target detection model based on the YOLO deep learning framework is used as the image recognition model for construction procedures. The dataset of the pre-trained target detection model is constructed by collecting process node images and process completion label images corresponding to the construction procedure standards.

[0022] The construction process image recognition model identifies the process completion nodes in the video stream data, counts the number of completed processes, calculates the progress completion rate of the completed processes in the current target construction area relative to the total processes, and simultaneously extracts the schedule plan of the remaining processes, which together serve as the construction progress data.

[0023] Preferably, extracting the remaining quantity data of each type of material in the fixed material storage area specifically includes:

[0024] The individual volume and unit volume weight parameters of the corresponding material category are pre-stored in the local database of the scheduling controller;

[0025] Identify the types of materials in the fixed material storage area and calculate the storage volume of the corresponding material type;

[0026] By combining the individual volume and unit volume weight parameters of the corresponding material category pre-stored in the local database of the scheduling controller, the remaining quantity data of the material category is calculated.

[0027] Preferably, when calculating the theoretical material requirements for completing the remaining construction procedures in the current target construction area, a dynamic material loss correction step is also included, specifically including:

[0028] Using a multiple linear regression model in industrial data analysis, with historical material and reagent loss data, construction personnel skill levels, and construction environment parameters of the same type of construction process as input, the dynamic material loss correction coefficient for the corresponding process is calculated.

[0029] The theoretical material demand is corrected by a dynamic material loss correction coefficient to obtain the corrected actual demand. The corrected actual demand is then matched and verified with the remaining quantity data to generate a material supply and demand matching result.

[0030] Preferably, it also includes a step for monitoring construction violations simultaneously:

[0031] The video stream data collected by the monitor is used to identify the compliance of construction workers' construction steps based on a pre-trained violation recognition model. When the compliance of construction steps is abnormal, a violation record is generated.

[0032] Violation records are linked to the corresponding construction personnel's unique identification code and stored in the local database of the dispatch controller.

[0033] Preferably, the dispatch controller generates material replenishment dispatch instructions for the corresponding category and quantity, specifically including:

[0034] Based on the material supply and demand matching results, combined with the construction progress data corresponding to the independent identification code bound to the target construction area, the schedule plan of the remaining processes, and the skill level information of the construction personnel, the urgent priority of material shortages is divided.

[0035] The system synchronously acquires inventory data of all material storage points in large-scale buildings, location and load status data of each material transfer unit, and on-site transfer path data generated based on the BIM model of the target construction area. It uses the highest priority of material shortage urgency, the lowest risk at the end of the process, the best transfer timeliness, and the lowest transfer cost as the multi-objective optimization function, and the corrected actual demand as the minimum loading constraint for a single transfer. It solves the optimal transfer path, transfer sequence, and loading allocation scheme through an improved multi-objective genetic algorithm in industrial data analysis, and generates corresponding material replenishment scheduling instructions based on the optimal scheme.

[0036] Preferably, the generation of cross-regional material dispatching instructions by the dispatch controller specifically includes:

[0037] For all target construction areas, based on the construction progress data, remaining construction procedures, and dynamic material loss correction coefficient corresponding to the bound independent identity code, calculate the safety reserve of the same type of materials in each target construction area, and determine the part of the remaining material quantity that exceeds the safety reserve as schedulable redundant material.

[0038] By using the global supply and demand balance optimization model in industrial data analysis, all schedulable redundant materials and material shortages are globally matched. The optimization objectives are: priority for materials from the same building and floor, priority for process sequence connection, shortest total transfer distance, and shortest total transfer period. The appearance integrity and specification matching degree of schedulable redundant materials are verified by video stream data collected by the monitor, and the globally optimal cross-regional material scheduling scheme is obtained.

[0039] Based on the optimal solution, corresponding cross-regional material scheduling instructions are generated. At the same time, the scheduling material information is associated with the independent identification code bound to the target construction area, and the remaining data of the corresponding standard material demand list is updated.

[0040] Preferably, it also includes a model self-optimization iteration step, including: collecting deviation data between each material supply and demand matching judgment result and actual material consumption data, verification data of the recognition results of the construction process image recognition model and the material category recognition and surplus calculation model, and misjudgment and omission data of the violation recognition model;

[0041] Associate the construction worker's skill level, the process type of the target construction area, and the apartment type characteristics with the corresponding independent identification code of the data;

[0042] When the deviation data or identification error data exceeds the preset threshold, incremental learning iteration is triggered, or all data is aggregated at fixed times every day for batch incremental learning iteration.

[0043] During the iteration process, the dataset was classified and trained according to the skill level of construction personnel, the type of work process, and the characteristics of the apartment type. The parameters of the construction process image recognition model, the material category recognition and surplus calculation model, the material demand calculation model, and the violation behavior recognition model were optimized respectively. The parameters of the iterated models were matched and adapted according to the classification labels, the corresponding target construction areas, and the independent identity codes.

[0044] Compared with existing technologies, this invention provides a method for dynamic scheduling and optimization of intelligent construction resources based on industrial data analysis, which has the following beneficial effects:

[0045] 1. By using a strong binding mechanism between the independent identification code of construction personnel and the target construction area and construction process, and with the monitoring device, the construction progress and material reserves are collected simultaneously from two dimensions. Combined with the industrial data analysis engine, the accurate prediction and matching verification of material supply and demand is completed. This breaks away from the passive mode of relying on manual feedback of material shortages in traditional indoor construction. It can identify the risk of material shortage in advance and trigger precise replenishment, avoiding the problem of work stoppage due to material shortage in the indoor construction of large-scale buildings. It greatly ensures the continuity of construction processes and effectively avoids the risk of chain delays in construction of separate units.

[0046] 2. Based on industrial data analysis, a multi-objective optimized material scheduling system was constructed. Through dynamic material loss correction coefficient, the actual material demand and safety reserve were accurately calculated. This system not only realized the priority division and optimal solution of material replenishment and transportation, but also completed the identification of redundant materials across the entire project and global cross-regional matching and scheduling. This achieved the global optimal flow and reuse of material resources, significantly reduced the cost of material backlog losses, duplicate purchases and ineffective transportation, and improved the comprehensive utilization efficiency of construction material resources.

[0047] 3. By collecting construction progress data and remaining data through the monitor, the system simultaneously identifies and warns of safety hazards such as violations of construction procedures and objects thrown from heights, reducing on-site management costs. At the same time, through the self-optimization mechanism of the classification incremental learning model, the identification and calculation model can be continuously iterated and optimized in combination with the actual construction data of the project, adapting to the personalized needs of different construction personnel, procedures and unit types, continuously improving scheduling accuracy and management capabilities, and possessing good scenario adaptability and long-term application value. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating the steps of the intelligent construction resource dynamic scheduling optimization method based on industrial data analysis in this application embodiment;

[0049] Figure 2 This is a flowchart illustrating the process of calculating construction progress data in the intelligent construction resource dynamic scheduling optimization method based on industrial data analysis, as described in this application embodiment.

[0050] Figure 3 This is a flowchart illustrating the process of extracting the remaining quantity data of each type of material in a fixed material storage area using the intelligent construction resource dynamic scheduling optimization method based on industrial data analysis, as described in this application embodiment.

[0051] Figure 4 This is a flowchart illustrating the specific implementation steps of the material theoretical demand correction and supply-demand matching verification in the intelligent construction resource dynamic scheduling optimization method based on industrial data analysis, as described in this application embodiment. Detailed Implementation

[0052] The following is in conjunction with the appendix Figure 1-4 This application will be described in further detail.

[0053] This application discloses a method for dynamic scheduling and optimization of intelligent construction resources based on industrial data analysis. (Refer to...) Figure 1 The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis includes:

[0054] S1. Assign a unique identification code to each construction worker involved in indoor construction, and bind the unique identification code to the target construction area, preset construction procedure, and standard material requirement list corresponding to the construction worker. This achieves a strong association between people, site, materials, and procedures, forming bound data. The bound data is stored in the local database of the scheduling controller, and the corresponding construction area and construction worker can be quickly matched through the local database.

[0055] S2. Each construction worker wears a monitor that is uniquely bound to their own unique identification code. The monitor's field of view is controlled to rotate synchronously with the construction worker's field of view, collect video data streams of the construction scene, and transmit the video stream data carrying the corresponding unique identification code to the scheduling controller to ensure that each video stream data can be accurately traced back to the corresponding construction worker and target construction area.

[0056] In this embodiment, each construction worker wears a monitor uniquely bound to an independent identification code to achieve real-time synchronous acquisition of video stream data at the construction site, eliminating blind spots in detection.

[0057] S3. The scheduling controller parses and processes the received video data stream, identifies the completed process nodes in the current target construction area based on the pre-trained construction process image recognition model, and calculates the construction progress data.

[0058] The monitor can be a follow-up visual monitor, used to realize the follow-up synchronous acquisition of video stream data at the construction site. It is an existing video stream data acquisition device and will not be described in detail here.

[0059] S4. When the monitor's field of view is detected to be directed toward the fixed material stacking area within the target construction area, the remaining quantity calculation process is triggered. Based on the pre-trained material category identification model and remaining quantity calculation model, the remaining quantity data of each category of material in the fixed material stacking area is extracted.

[0060] S5. The scheduling controller takes the standard material requirement list and construction progress data as input, calculates the theoretical material requirement required to complete the remaining construction procedures in the current target construction area, matches and verifies the theoretical material requirement with the remaining quantity data of the corresponding material category, and generates a material supply and demand matching judgment result.

[0061] S6. Based on the material supply and demand matching determination result, if the material supply and demand matching determination result is that the current remaining material quantity is insufficient to support the completion of the remaining construction process, the scheduling controller generates a material replenishment scheduling instruction for the corresponding category and quantity, and transfers the required material to the fixed material stacking area of ​​the target construction area through the material transfer unit; if it is identified that there are redundant materials of the same type in the target construction area that exceed the demand of its remaining processes, the scheduling controller generates a cross-regional material scheduling instruction and transfers the redundant material to the target construction area with insufficient remaining material quantity.

[0062] Furthermore, each construction worker involved in indoor construction is assigned a unique identification code, which is then linked to the worker's target construction area, pre-defined construction procedures, and standard material requirements list. Specifically, this includes:

[0063] S11. Based on the BIM model of the target construction area, the schedule plan of the preset construction process, and the material consumption quota of the unit process, calculate the standard material usage of a single process and the total usage of the whole process, and generate the standard material requirement list for the corresponding target construction area.

[0064] S12. Synchronously bind and store the skill level information and safety training record information of the corresponding construction personnel with their independent identification codes.

[0065] The BIM model corresponds to the house size and work area of ​​a single household.

[0066] Before construction workers enter the site, it is necessary to collect their skill level information and safety training records.

[0067] The skill level information is divided into three levels: primary, intermediate, and advanced, based on the construction workers' years of service and assessment results. The safety training record information includes the training certificate number, training content, training time, and validity period. All information is entered into the dispatch controller after being verified by the project safety management department.

[0068] Furthermore, the verified skill level information and safety training records of construction personnel will be synchronously bound to the unique identification code of the construction personnel. During the binding process, a relationship will be established through the unique identification code to ensure that each piece of personnel information can be accurately matched with the corresponding construction personnel and target construction area.

[0069] For example, a construction worker with an independent identification code 05-18-03-01, whose skill level is intermediate and whose safety training certificate number is AQ-2025001, will have these three pieces of information bound to the code 05-18-03-01 and stored synchronously in the local database of the dispatch controller.

[0070] Furthermore, when construction workers upgrade their skill levels, renew their safety training records, or change their target construction area, project managers promptly update the relevant information in the project management system. The updated information is then synchronized to the local database of the dispatch controller to ensure the reliability and accuracy of the bound data.

[0071] Meanwhile, the video data streams collected from the construction site specifically include:

[0072] S21. Collect head posture data of construction workers through the monitor, and drive the monitor to adjust the shooting angle synchronously with the head rotation of the construction workers based on the head posture data, so that the image collected by the monitor is synchronously matched with the field of vision of the construction workers.

[0073] S22. The monitor adds a unique identifier to the collected video stream data and transmits the identified video stream data to the scheduling controller via a wireless local area network.

[0074] The first step, hardware deployment and parameter initialization, involves using a head-mounted binocular vision camera to monitor the workers. This camera is fixed in the center of the front of the worker's safety helmet, parallel to the worker's eye level. The monitor has a built-in six-axis MEMS attitude sensor. Before the workers enter the site, the sensor's zero-point calibration and lens distortion correction are performed to eliminate acquisition errors caused by installation deviations and lens distortion. At the same time, the sensor's sampling frequency is initialized to 100Hz, the monitor's acquisition frame rate is 30fps, and the image resolution is 1920×1080 to ensure that the clarity of the acquired image and the accuracy of the attitude data meet the requirements of on-site operations.

[0075] The second step is head posture data acquisition and calculation: During the construction work, the six-axis MEMS attitude sensor collects the raw data of the construction worker's three-axis angular velocity and three-axis acceleration at a preset sampling frequency of 100Hz. The raw data is filtered and denoised by the built-in Kalman filter algorithm to calculate the attitude data of the construction worker's head, including three core parameters: yaw angle, pitch angle and roll angle. This accurately reflects the rotation direction and viewing angle of the construction worker's head, with a calculation accuracy error of ≤0.5°.

[0076] The third step is to synchronize the camera's shooting angle with the field of view: The calculated head posture data is synchronously transmitted to the monitor's pan-tilt control unit. Based on the head posture parameters, the pan-tilt control unit synchronously drives the monitor's horizontal and vertical rotation mechanisms to adjust the shooting angle, so that the monitor's shooting angle is synchronously matched with the construction worker's line of sight in a 1:1 ratio, and the rotation response delay is controlled within 50ms. Specifically, when the construction worker's head turns towards the work surface, the monitor synchronously aims at the work surface to collect images related to the construction progress. When the construction worker's head turns towards the fixed material stacking area within the target construction area, the monitor synchronously aims at the material stacking area to collect images related to the remaining material, ensuring that the collected video stream data completely covers the construction worker's actual field of view and completely eliminates the coverage blind spots of the fixed monitoring equipment.

[0077] In this process, based on the acquisition of video stream images, a unique identifier is added to each video stream data segment, which is bound to the construction personnel. This ensures that the dispatch controller can accurately trace the construction personnel and target construction area corresponding to each frame, avoiding confusion of video stream data from multiple workers. The specific implementation steps are as follows:

[0078] Pre-writing and association of unique identifiers: The monitor has a built-in UHF radio frequency identification tag that is uniquely bound to the corresponding construction personnel's unique identification code. The tag is pre-written with the construction personnel's complete unique identification code. The unique identification code is completely consistent with the bound target construction area, preset construction procedures, and standard material requirement list, ensuring the uniqueness, relevance, and traceability of the identification.

[0079] Adding frame-level identifiers to video stream data: The monitor's encoding processing unit performs frame-by-frame hard encoding on the acquired video stream data. During the encoding process, a pre-stored independent identification code is written into the SEI supplementary enhancement information field of each video frame as a unique identifier. At the same time, the acquisition timestamp of that frame is also written to ensure that each frame of video data can be accurately matched with the corresponding construction personnel and acquisition time, thereby fundamentally solving the problem of chaotic matching of video stream data from multiple construction areas and multiple workers.

[0080] Wireless transmission of video stream data: A fully covered 5G wireless LAN is pre-built on the project site, and wireless AP nodes are deployed on each floor of each residential building to ensure full network signal coverage in all indoor construction areas, with a signal strength ≥ -65dBm and a transmission bandwidth ≥ 100Mbps; the monitor uploads video stream data with a unique identifier to the dispatch controller in a low-latency transmission mode through the built-in WiFi 6 wireless module, with end-to-end transmission latency controlled within 200ms; at the same time, a network interruption resumption mechanism is set up. When the network is temporarily interrupted, the monitor caches the video stream data locally to the built-in storage chip, and automatically re-transmits the cached data after the network is restored, ensuring the integrity of the video stream data.

[0081] It should be noted here that when collecting video data streams of the construction scene through the monitor, the acquisition cycle of the video stream data can be changed according to actual needs, such as periodic acquisition, non-periodic acquisition, or real-time acquisition. This can be changed according to the computing power and performance of the scheduling controller, and is not limited here.

[0082] Furthermore, refer to Figure 2 The calculated construction progress data specifically includes:

[0083] S31. A target detection model pre-trained based on the YOLO deep learning framework is used as a construction process image recognition model. The dataset of the pre-trained target detection model is constructed by collecting process node images and process completion label images corresponding to the construction process standards.

[0084] S32. Identify the process completion nodes in the video stream data through the construction process image recognition model, count the number of completed processes, calculate the progress completion rate of the completed processes in the current target construction area relative to the total processes, and simultaneously extract the schedule plan of the remaining processes, which together serve as the construction progress data.

[0085] include:

[0086] S311. Construction and preprocessing of special datasets: For the bound preset construction procedures, collect several real-scene images of indoor construction scenes of large-scale buildings to construct a special dataset. The images cover different apartment types, different lighting conditions, different working angles of construction workers, and different completion levels of the procedures. The dataset includes standard images of completed nodes as positive samples and images of incomplete nodes and interference scenes as negative samples.

[0087] Meanwhile, the LabelImg annotation tool is used to standardize the annotation of all images. The annotation content includes process category, node completion status, and key visual features. The annotation results correspond one-to-one with the standardized decomposition nodes of the preset construction process, and the annotation accuracy error is ≤1 pixel.

[0088] S312. Preprocess and enhance the labeled dataset: First, uniformly change the image resolution of all images to 640×640 and normalize the pixel values ​​to eliminate the impact of size differences on model training; then, expand the dataset through data augmentation methods such as random flipping, brightness / contrast adjustment, viewpoint distortion simulation, and local occlusion simulation to solve the recognition error problems caused by changes in lighting, viewpoint shaking, and partial occlusion of materials in construction scenes; finally, divide the enhanced dataset into training set, validation set, and test set in an 8:1:1 ratio to ensure that the training, validation, and test scenarios do not overlap and to guarantee the generalization ability of the model.

[0089] S313. A target detection model pre-trained based on the YOLOv8 deep learning framework is adopted as the image recognition model for construction procedures. This model balances inference speed and recognition accuracy, and is adapted to the scheduling controller's requirements for video stream parsing. For the visual features of indoor construction procedures, the model is specifically adapted and optimized: anchor box sizes adapted to the features of the procedures are generated by re-clustering based on the specific dataset, which enhances the model's ability to extract texture and flatness features of large areas such as walls and floors. The CIoU loss function is used to optimize the localization accuracy of the procedure area, and classification weights are introduced to solve the problem of imbalance of samples in different procedures.

[0090] S314. Use transfer learning to complete model training: first, pre-train the model based on the special dataset, and then fine-tune the training based on the special dataset.

[0091] The trained model is pruned and INT8 quantized to reduce its size to less than 6MB, and the single-frame image inference speed is increased to ≤10ms. It can be adapted to the edge computing power of the scheduling controller and meet the parsing requirements of 30fps video stream.

[0092] Furthermore, refer to Figure 3 Extracting the remaining quantity data of each type of material in the fixed material storage area specifically includes:

[0093] The individual volume and unit volume weight parameters of the corresponding material category are pre-stored in the local database of the scheduling controller;

[0094] Identify the types of materials in the fixed material storage area and calculate the storage volume of the corresponding material type;

[0095] By combining the individual volume and unit volume weight parameters of the corresponding material category pre-stored in the local database of the scheduling controller, the remaining quantity data of the material category is calculated.

[0096] The specific implementation process includes:

[0097] S41. Pre-storage and association binding steps for basic material parameters:

[0098] S411. Parameter Extraction and Standardization: Based on the generated standard material requirement list for the target construction area, extract the basic parameters of all categories of materials in the list. For block materials, extract the length, width, and thickness parameters of a single block and calculate the volume of a single piece. For powder materials, extract the industry-standard unit volume weight parameters. After all parameters are verified to be correct, complete the standardization.

[0099] S412. Parameter Pre-storage and Association Binding: The individual volume and unit volume weight parameters of the corresponding material category are associated and bound with the standard material requirement list to which the material belongs and the independent identification code of the corresponding target construction area, and pre-stored in the local database of the scheduling controller. The storage format adopts a structured table that corresponds one-to-one with the standard material requirement list, including five core fields: material category, specification, individual volume, unit volume weight, corresponding independent identification code, and target construction area. This facilitates quick retrieval and matching during surplus calculation and avoids parameter confusion between different regions and different categories.

[0100] S413. Parameter Update and Maintenance: When the material categories and specifications in the standard material requirements list change, the project material management personnel will update the basic parameters of the corresponding materials simultaneously. The updated parameters will be synchronized to the local database of the scheduling controller to ensure that the pre-stored parameters are completely consistent with the materials actually used on site, providing an accurate benchmark for subsequent reserve calculation.

[0101] S42. Specific implementation steps for material category identification and stacking volume calculation:

[0102] S421, Triggering of Calculation Process and Extraction of Effective Images: When the dispatch controller recognizes that the field of view of the monitor is facing the fixed material stacking area in the target construction area, it automatically triggers the margin calculation process; it retrieves the current frame and two consecutive clear frames collected by the monitor, and combines them with the head posture data of the monitor to filter out effective images that are unobstructed and free from strong light overexposure, as the input source for subsequent identification and calculation.

[0103] S422. Accurate Material Category Identification: Based on a pre-trained material category identification model, the effective images are identified and classified. This model uses the ResNet50 image classification network, which is a well-known and mature model in the field of image classification. In this embodiment, only the indoor construction material scene has been fine-tuned and optimized. The pre-training dataset contains real-world images of all types of materials used in the project under different stacking states and different lighting conditions. The classification and identification accuracy is no less than 98%. After the identification is completed, the category and coordinates of each material in the image are output to define the effective calculation range for subsequent volume calculation.

[0104] S423. Accurate Calculation of Material Stacking Volume: For the identified and delineated effective area of ​​a single type of material, a binocular visual ranging and 3D point cloud reconstruction algorithm (binocular stereo vision 3D reconstruction is a well-known and mature technology; in this embodiment, point cloud denoising and stacking contour fitting logic are optimized for indoor construction scenarios) are used. First, the depth information of each point of the material stack is obtained through binocular parallax calculation, and a 3D point cloud model of the material stack is generated. Then, invalid point cloud data such as packaging debris and ground interference are removed, and the effective contour of the material stack is fitted to calculate the actual stacking volume of the corresponding type of material. The calculation error is ≤3%.

[0105] S43. Steps for converting and updating residual data:

[0106] S431. Material Remaining Amount Conversion by Type: For different types of materials, the remaining amount is calculated using the corresponding conversion rules. For block materials, the remaining amount is calculated using the formula: Remaining Amount = Material Stacking Volume / Individual Volume, based on the pre-stored unit volume parameters. For powder materials, the remaining weight is calculated using the formula: Remaining Amount = Material Stacking Volume × Unit Volume Weight, based on the pre-stored unit volume weight parameters.

[0107] For example, within the target construction area with an independent identification code of 05-18-03-01, the calculated volume of stacked tiles is 0.32m³. 3 The pre-stored volume of a single tile is 0.00064 m³. 3 Therefore, the remaining number of tiles is 0.32 / 0.00064 = 500 tiles;

[0108] S432. Data Binding and Storage: Bind the calculated remaining quantity data of each type of material to the independent identification code of the corresponding material category, the target construction area, and the calculation timestamp, and store it synchronously in the local database of the scheduling controller. At the same time, update the remaining quantity field of the corresponding standard material requirement list to ensure that the latest remaining quantity data can be directly retrieved when calculating material requirements.

[0109] S433, Outlier Validation and Correction: For the converted remaining quantity data, outlier validation is performed by combining the historical consumption data of the region with the total usage of the standard material requirement list. If the remaining quantity data exceeds a reasonable threshold, the calculation process is retried to avoid calculation errors caused by screen occlusion or point cloud distortion, and to ensure the accuracy of the remaining quantity data.

[0110] Furthermore, when calculating the theoretical material requirements for completing the remaining construction procedures in the current target construction area, a dynamic material loss correction step is also included, specifically:

[0111] Using a multiple linear regression model in industrial data analysis, with historical material and reagent loss data, construction personnel skill levels, and construction environment parameters of the same type of construction process as input, the dynamic material loss correction coefficient for the corresponding process is calculated.

[0112] The theoretical material demand is corrected by a dynamic material loss correction coefficient to obtain the corrected actual demand. The corrected actual demand is then matched and verified with the remaining quantity data to generate a material supply and demand matching result.

[0113] Specifically, it includes:

[0114] A1. Model Input / Output Variables and Baseline Dataset Construction: Define the model's input and output variables, with input variables including four core dimensions:

[0115] Firstly, historical data on actual material losses during similar construction procedures;

[0116] Secondly, information on the skill levels of construction workers;

[0117] Third, construction environment parameters;

[0118] Fourth, process type;

[0119] Furthermore, the output variable is the actual material loss rate of the corresponding process, which is ultimately converted into a dynamic material loss correction coefficient.

[0120] Meanwhile, based on the above variables, a benchmark dataset containing valid historical data was constructed. After removing outliers and missing values, the data was standardized and divided into training, validation, and test sets in a ratio of 7:2:1 to provide data support for model training.

[0121] A11. Model Training and Significance Validation: Based on the constructed benchmark dataset, the multiple linear regression model is trained and fitted. The regression coefficients of the model are solved using the well-known least squares method. The training process is a common and mature process for multiple linear regression, which will not be elaborated in this embodiment. After training, the F-test is used to verify the overall significance of the model, and the t-test is used to verify the significance of individual input variables to ensure that all input variables have a significant impact on the material loss rate. Insignificant variables are removed to optimize the model fitting effect. The final trained model has a determination coefficient R² ≥ 0.92 on the test set, which meets the requirements for on-site prediction accuracy.

[0122] A12. Model Deployment and Update: Deploy the trained dynamic material loss correction coefficient prediction model to the industrial data analysis engine of the scheduling controller, set up a monthly automatic update mechanism, summarize the actual material loss data of the project each month, supplement it to the benchmark dataset, and perform incremental fitting optimization on the model to ensure that the model continuously adapts to the actual construction characteristics of the project.

[0123] A2. Specific implementation steps for revising theoretical material requirements and verifying supply and demand matching, refer to... Figure 4 ,include:

[0124] A21. Basic calculation of the theoretical material requirements for remaining processes: The scheduling controller calls the industrial data analysis engine, taking the generated standard material requirements list and the calculated construction progress data as input, firstly to obtain the remaining unfinished construction processes in the current target construction area, and then, combined with the standard material usage of the corresponding process in the standard material requirements list, to calculate the theoretical material requirements required to complete the remaining processes. The calculation formula is: Theoretical material requirements = Sum of standard usage of each remaining process × Remaining process quantity.

[0125] A22. Calculation of dynamic material loss correction coefficient: For the current remaining construction process, retrieve the corresponding input variable data, including historical loss data of the same type of process, skill level information of the construction personnel bound to the process, indoor temperature and humidity and base flatness parameters of the current target construction area, and process type. Input the above data into the trained multiple linear regression model. The model outputs the predicted material loss rate corresponding to the process, which is the dynamic material loss correction coefficient.

[0126] For example, if the predicted loss rate of the wall leveling process for a construction worker with intermediate skill level is 8%, then the dynamic material loss correction coefficient for this process is 0.08.

[0127] A23. Correction of theoretical material requirements: The theoretical material requirements obtained from the basic calculation are corrected by the dynamic material loss correction coefficient to obtain the corrected actual requirements. The correction calculation formula is: Corrected actual requirements = theoretical material requirements × (1 + dynamic material loss correction coefficient). This ensures that the calculated material requirements are fully matched with the actual loss characteristics of the current construction personnel and construction environment, and avoids the calculation deviation caused by the fixed loss coefficient.

[0128] A24. Material Supply and Demand Matching Verification and Judgment Result Generation: The corrected actual demand quantity is matched and verified with the calculated remaining quantity data of the corresponding material category. If the remaining quantity is greater than or equal to the corrected actual demand quantity, a material supply and demand matching judgment result is generated indicating that the remaining material quantity can support the completion of the remaining processes. If the remaining quantity is less than the corrected actual demand quantity, a material supply and demand matching judgment result is generated indicating that the remaining material quantity is insufficient to support the completion of the remaining processes. The judgment result is bound to the independent identity code of the corresponding target construction area and stored in the local database of the scheduling controller.

[0129] Furthermore, it also includes a step for monitoring construction violations simultaneously:

[0130] The video stream data collected by the monitor is used to identify the compliance of construction workers' construction steps based on a pre-trained violation recognition model. When the compliance of construction steps is abnormal, a violation record is generated.

[0131] Violation records are linked to the corresponding construction personnel's unique identification code and stored in the local database of the dispatch controller.

[0132] Specifically, it includes:

[0133] B1. Construction and Deployment Steps of the Pre-trained Violation Recognition Model: The construction and deployment steps of the pre-trained violation recognition model provide stable and accurate algorithmic support for the compliance recognition of construction steps. A computer vision model is used to complete the adaptation to specific scenarios. The specific implementation steps are as follows:

[0134] B11. Construction and Annotation of Specialized Datasets: A specialized dataset for violations related to indoor construction projects will be constructed. This dataset includes a collection of real-world images covering two core violation scenarios:

[0135] Firstly, scenarios involving violations of procedures and steps;

[0136] Secondly, violations of safe operating procedures, such as throwing objects from heights or failing to wear safety helmets / safety belts and other protective equipment as required;

[0137] Meanwhile, the LabelImg tool was used to standardize the annotation of all images. The annotation content included the violation type, key features of the violation, and corresponding process category. The annotation accuracy error was ≤1 pixel. The images were divided into training set, validation set, and test set in a ratio of 8:1:1.

[0138] B12. Model Selection, Adaptation, and Training: An action recognition model based on the YOLOv8 deep learning framework is adopted as the violation recognition model. This model is a well-known and mature action recognition framework in the field of computer vision. This embodiment only performs special adaptation and optimization for indoor construction violation scenarios, and strengthens the feature extraction capability for minor process violations and small target safety protection equipment. The model is trained using transfer learning, and the recognition accuracy on the test set after training is not less than 98.5%.

[0139] B13. Model Lightweighting and Deployment: The trained model is pruned and quantized to reduce its size and adapt to the edge computing power of the scheduling controller. The single-frame inference speed is ≤12ms to meet the requirements of video stream parsing. The optimized model is deployed to the local inference unit of the scheduling controller and shares the same inference computing power with the construction process image recognition model and the material category recognition model, without the need for additional hardware deployment.

[0140] B2. Construction procedure compliance identification and violation record generation steps:

[0141] B21. Synchronous retrieval and preprocessing of video stream data: While the scheduling controller analyzes the construction progress and material balance of the video stream data, it simultaneously diverts the video stream data to the violation recognition model, extracts key frames at a frequency of 3 frames per second, and inputs them into the violation recognition model after standardized preprocessing. The preprocessing process is a commonly used technique in the field of image recognition, which will not be described in detail in this embodiment.

[0142] B22. Construction Step Compliance Identification: Based on a pre-trained violation identification model, the model performs two-dimensional compliance identification on the input keyframes: First, it identifies the compliance of the construction steps, verifying whether the construction personnel's work steps conform to the preset construction procedure standards and specifications, and whether there are violations such as reversed or omitted procedures; second, it identifies the compliance of safe operation, verifying whether the construction personnel have violated regulations such as throwing objects from heights or not wearing safety protective equipment as required. The model outputs the violation type and confidence level of the identification results, and only when the confidence level is ≥90% is it judged as a valid identification result.

[0143] B23. Anomaly Detection and Violation Record Generation: When an anomaly is detected in the compliance of a construction step, i.e., a valid violation identification result exists, the violation record generation process is immediately triggered. The violation record includes core fields: violation timestamp, violation type, violation content description, violation key frame video clip, the independent identification code of the corresponding construction personnel, and the information of the target construction area, ensuring that each violation record can be accurately traced and verified.

[0144] B3. Steps for binding, storing, and managing violation records in a closed loop:

[0145] B31. Binding of violation records to independent identification codes: Based on the unique identifier carried at the frame level of video stream data, the independent identification code corresponding to the violation scene is extracted, and the generated violation record is uniquely bound to the independent identification code to ensure that each violation record is accurately matched with a specific construction worker, avoiding confusion of violation records of multiple personnel and multiple construction areas;

[0146] B32. Storage and Ledger Management of Violation Records: Completed violation records are stored in the local database of the dispatch controller and archived in conjunction with the bound construction personnel's skill level information and safety training records to form an electronic ledger for project construction violation management. The ledger supports multi-dimensional queries by independent identity code, construction area, violation type, and violation time, making it convenient for project management personnel to retrieve and review.

[0147] B33. Violation Warning and Closed-Loop Handling: When a violation record is generated, the dispatch controller simultaneously sends a warning to the handheld terminal of the on-site management personnel and the monitor of the corresponding construction personnel, urging the site to rectify immediately; for construction personnel whose cumulative number of violations exceeds the preset threshold, a safety training review reminder is triggered simultaneously, forming a control closed loop with the safety training record.

[0148] Furthermore, the dispatch controller generates material replenishment dispatch instructions for the corresponding category and quantity, specifically including:

[0149] Based on the material supply and demand matching results, combined with the construction progress data corresponding to the independent identification code bound to the target construction area, the schedule plan of the remaining processes, and the skill level information of the construction personnel, the urgent priority of material shortages is divided.

[0150] The system synchronously acquires inventory data of all material storage points in large-scale buildings, location and load status data of each material transfer unit, and on-site transfer path data generated based on the BIM model of the target construction area. It uses the highest priority of material shortage urgency, the lowest risk at the end of the process, the best transfer timeliness, and the lowest transfer cost as the multi-objective optimization function, and the corrected actual demand as the minimum loading constraint for a single transfer. It solves the optimal transfer path, transfer sequence, and loading allocation scheme through an improved multi-objective genetic algorithm in industrial data analysis, and generates corresponding material replenishment scheduling instructions based on the optimal scheme.

[0151] Specifically, it includes:

[0152] S61. Specific implementation steps for prioritizing material shortages:

[0153] S611, Core Related Data Retrieval: When the remaining amount of generated materials is insufficient to support the material supply and demand matching judgment result for the completion of the remaining processes, the scheduling controller immediately retrieves the related data of the target construction area corresponding to the judgment result, including the construction progress data corresponding to the independent identity code bound to the area, the schedule plan of the remaining processes, the skill level information of the construction personnel, and the category and quantity of material shortages, providing data support for priority division.

[0154] S612. Priority Grading Rules: Based on the urgency of the remaining work processes and combined with the skill level of the construction personnel, priority is assigned in four levels from high to low: Special Priority, Level 1, Level 2, and Level 3. The rules are as follows: Material shortages with less than 24 hours remaining are directly classified as Special Priority; material shortages with 24-72 hours remaining are classified as Level 1 Priority; material shortages with 72-168 hours remaining are classified as Level 2 Priority; and material shortages with more than 168 hours remaining are classified as Level 3 Priority. For work processes handled by construction personnel with basic skill levels, the priority of their material shortages is automatically increased by one level to further reduce the risk of work process interruption.

[0155] S63. Priority Result Binding and Storage: The emergency priority of the material shortage is bound to the unique identification code, material category, and shortage quantity of the corresponding target construction area, and stored in the local database of the scheduling controller as the core input parameter for subsequent multi-objective optimization solution.

[0156] S62. Steps for constructing a multi-objective optimization model and solving for the optimal transport scheme:

[0157] The steps for constructing the multi-objective optimization model and finding the optimal transport scheme are based on the highest priority constraint. A multi-objective optimization function is constructed, and an improved multi-objective genetic algorithm is used to find the globally optimal transport scheme. The specific implementation steps are as follows:

[0158] S621. Synchronous Acquisition of Basic Transfer Data: The dispatch controller synchronously acquires three types of core basic data: First, inventory data of all material storage points in the large-scale building, including the corresponding material categories, inventory quantities, and locations of each storage point; second, location, empty / loaded status, maximum load, and travel speed data of each material transfer unit within the project; and third, on-site transfer path data generated based on the BIM model of the target construction area, including information such as the location of freight elevators within the building, floor passages, indoor access routes, and path access restrictions, while eliminating invalid paths occupied by construction fences and obstacles.

[0159] S622. Construction of Multi-Objective Optimization Function and Constraints: The multi-objective optimization function is based on the following: highest priority for material shortage urgency, lowest risk of process interruption, optimal transfer timeliness, and lowest transfer cost. The four objectives are normalized and weighted, with the urgency priority having the highest weight. At the same time, core constraints are set, with the obtained modified actual demand as the minimum loading constraint for a single transfer. The maximum load constraint of the transfer unit, the path access constraint, and the material delivery timeliness constraint are also set to ensure that the solution meets the actual on-site operation requirements.

[0160] S623. Optimal Solution Calculation: An improved multi-objective genetic algorithm is used to find the optimal solution. The improved multi-objective genetic algorithm is based on the NSGA-II algorithm. The improved multi-objective genetic algorithm is a well-known and mature algorithm in the field of multi-objective optimization. The general solution process will not be described in this embodiment. Specific improvements are made for this indoor construction scenario. During the population iteration process, the optimal solution of high-priority gaps is retained first. At the same time, dynamic obstacle avoidance constraints of BIM paths are introduced. Finally, the optimal transfer path, transfer sequence, and loading allocation scheme that simultaneously meet the requirements of multiple objectives are obtained, ensuring that high-priority gaps are delivered first and materials on the same path are loaded in a concentrated manner, minimizing transfer costs and transfer time.

[0161] S63. Steps for generating and executing material replenishment and dispatching instructions:

[0162] Specifically, the generation and closed-loop execution of material replenishment scheduling instructions are based on the optimal transfer scheme obtained from the solution. Standardized scheduling instructions are generated and issued for execution, while the entire process data closure is completed after execution. The specific implementation steps are as follows:

[0163] S631. Scheduling Instruction Generation and Binding: Based on the optimal transfer plan, standardized material replenishment scheduling instructions are generated. The instructions include core fields: material category, specifications, transfer quantity, start and end points, transfer route, transfer sequence, responsible transfer unit number, and unique identification code of the corresponding target construction area. After the instructions are generated, they are bound to the corresponding material supply and demand matching judgment result and unique identification code, and stored in the local database of the scheduling controller to achieve full-process traceability.

[0164] S632, Instruction Issuance and Execution: The dispatch controller will issue the generated material replenishment dispatch instructions to the vehicle terminal of the corresponding material transfer unit and the management terminal of the project material management department. The transfer unit driver will complete the material loading, transfer and delivery according to the instructions.

[0165] S633. Execution of closed loop and data update: After the material replenishment scheduling instruction is completed, the scheduling controller immediately triggers the material surplus update process of the corresponding target construction area, synchronously updates the surplus data and the surplus field of the standard material requirement list, and records the instruction execution time and completion status to form a closed loop ledger of scheduling instruction execution.

[0166] Furthermore, the specific steps by which the scheduling controller generates cross-regional material scheduling instructions include:

[0167] For all target construction areas, based on the construction progress data, remaining construction procedures, and dynamic material loss correction coefficient corresponding to the bound independent identity code, calculate the safety reserve of the same type of materials in each target construction area, and determine the part of the remaining material quantity that exceeds the safety reserve as schedulable redundant material.

[0168] By using the global supply and demand balance optimization model in industrial data analysis, all schedulable redundant materials and material shortages are globally matched. The optimization objectives are: priority for materials from the same building and floor, priority for process sequence connection, shortest total transfer distance, and shortest total transfer period. The appearance integrity and specification matching degree of schedulable redundant materials are verified by video stream data collected by the monitor, and the globally optimal cross-regional material scheduling scheme is obtained.

[0169] Based on the optimal solution, corresponding cross-regional material scheduling instructions are generated. At the same time, the scheduling material information is associated with the independent identification code bound to the target construction area, and the remaining data of the corresponding standard material demand list is updated.

[0170] The specific implementation steps of the above technical solution are as follows:

[0171] Accurate determination steps for scheduling redundant materials:

[0172] The precise determination of dispatchable redundant materials is based on construction and material data, combined with dynamic material loss characteristics to calculate the safety reserve, accurately defining the boundary of dispatchable redundant materials, and avoiding the problem of material shortages after being moved due to traditional rough determination. The specific implementation steps are as follows:

[0173] Synchronous retrieval of core data for the entire project: The scheduling controller synchronously retrieves all associated data of all target construction areas at fixed time intervals, including construction progress data corresponding to the independent identity code bound to each area, remaining construction procedures, dynamic material loss correction coefficients, and remaining quantity data of each type of material, providing full data support for the determination of redundant materials.

[0174] Precise calculation of safety reserve for single-category materials in a single area: For each type of material in each target construction area, based on the retrieved related data, the safety reserve for that type of material in that area is calculated. The calculation formula is as follows:

[0175] Safety reserve = Theoretical material requirement of remaining processes × (1 + Dynamic material loss correction coefficient) × 1.1, where 1.1 is the safety redundancy coefficient to adapt to unforeseen circumstances on site; the safety reserve is the minimum material inventory to ensure the smooth completion of the remaining processes in this area, thereby avoiding the risk of material shortages after cross-regional scheduling from the root.

[0176] Final determination of dispatchable redundant materials: The remaining quantity of the corresponding category of materials is compared with the calculated safety reserve quantity. Only when the remaining quantity exceeds the safety reserve quantity is the excess portion determined to be dispatchable redundant material. If the remaining quantity is less than or equal to the safety reserve quantity, there is no available redundancy for that category of materials in that area, and it is not included in the cross-regional dispatch scope. After the determination is completed, the category, quantity, location, and corresponding unique identification code of the dispatchable redundant materials are bound to form a project dispatchable redundant material ledger, which is stored in the local database of the dispatch controller.

[0177] For example, for a target construction area with an independent identification code of 05-18-04-02, the theoretical demand for the remaining tile process is 300 tiles, the dynamic material loss correction coefficient is 0.06, the safety reserve is 300×(1+0.06)×1.1≈350 tiles, and the remaining quantity is 500 tiles. Therefore, the schedulable redundancy of tiles in this area is 500-350=150 tiles.

[0178] Steps for solving the global supply and demand matching and optimal cross-regional scheduling scheme:

[0179] The steps for solving the global supply and demand matching and optimal cross-regional scheduling scheme are based on the schedulable redundant material ledger and the material shortage data of the entire project. A global supply and demand balance optimization model adapted to the indoor unit construction scenario is constructed. Combined with material compliance verification, the global optimal scheduling scheme is solved. The specific implementation steps are as follows:

[0180] Global supply and demand matching basic data preparation: The scheduling controller synchronously summarizes the material shortage data of the entire project, including the type, quantity, target construction area, remaining process period, and corresponding unique identification code of the shortage materials, and forms a basic database for supply and demand matching with the schedulable redundant material ledger; at the same time, it retrieves the site-wide transfer path data and access restriction information of each building floor generated by the project BIM model to provide a spatial benchmark for transfer path optimization;

[0181] Global supply and demand balance optimization model construction: Through the global supply and demand balance optimization model in industrial data analysis, all schedulable redundant materials and material shortages are globally matched. The model is based on linear programming and is specifically adapted for indoor unit construction scenarios. Multiple optimization objectives are set, with priorities from high to low as follows: priority for the same floor in the same building, priority for the connection of work sequence, shortest total transfer distance, and shortest total transfer period. This ensures that the scheduling plan prioritizes solving the supply and demand gap within the same building, minimizes long-distance transfers across buildings, and prioritizes the needs of work processes with tight schedules.

[0182] Simultaneously perform compliance verification of schedulable redundant materials: During the supply and demand matching process, the appearance integrity and specification matching degree of schedulable redundant materials are verified by the latest video stream data collected by the monitors in the corresponding target construction area. Invalid materials with damage, mismatched specifications and gaps are eliminated, and only compliant and usable redundant materials are included in the matching range to avoid invalid transfer and improve the material reuse rate.

[0183] Solution for the globally optimal scheduling scheme: Based on the constructed optimization model and the schedulable material list after compliance verification, the globally optimal cross-regional material scheduling scheme is obtained. The originating region, originating region, transfer quantity, transfer path, transfer sequence, and corresponding material category of each redundant material are clearly defined to ensure the global balance of material supply and demand for the entire project, while reducing transfer costs and transfer time.

[0184] Steps for generating cross-regional material dispatching instructions and updating data in a closed loop:

[0185] The cross-regional material scheduling instruction generation and data closed-loop update steps are based on the optimal scheduling scheme obtained from the solution. They generate standardized scheduling instructions and complete the closed-loop management of the entire process, while simultaneously updating the material data of the entire project to ensure data consistency. The specific implementation steps are as follows:

[0186] Cross-regional material dispatching instruction generation and binding: Based on the globally optimal cross-regional material dispatching scheme, standardized cross-regional material dispatching instructions are generated. The instructions include core fields: material category, specifications, transfer quantity, target construction area from which the material is dispatched, target construction area to which the material is dispatched, transfer route, transfer sequence, responsible transfer unit number, and independent identification code corresponding to the dispatching and receiving areas. After the instructions are generated, they are bound to the corresponding supply and demand matching data and independent identification code, and stored in the local database of the dispatch controller to achieve full-process traceability.

[0187] Instruction Issuance and Execution: The dispatch controller will simultaneously issue the cross-regional material dispatch instructions to the vehicle terminals of the corresponding material transfer units, the handheld terminals of the on-site management personnel in the sending and receiving areas, and the terminals of the project material management department; the transfer units will complete the sending out counting, transfer and receiving of materials according to the instructions. During the receiving process, the type, specifications and quantity of materials must be checked again to ensure that they are consistent with the instructions.

[0188] End-to-end data closed-loop update: After the receipt confirmation is completed, the dispatch controller immediately updates the material data of the two target construction areas: for the outgoing area, the corresponding number of dispatchable redundant materials is deducted and the remaining quantity data is updated; for the incoming area, the corresponding number of materials is added and the remaining quantity data is updated; at the same time, the dispatched material information is associated with the independent identification code bound to the receiving target construction area, and the remaining quantity field of the standard material requirement list is updated to form a complete data closed loop.

[0189] Furthermore, it also includes model self-optimization iteration steps, including: collecting deviation data between each material supply and demand matching judgment result and actual material consumption data, verification data of the recognition results of the construction process image recognition model and the material category recognition and surplus calculation model, and misjudgment and omission data of the violation recognition model.

[0190] Associate the construction worker's skill level, the process type of the target construction area, and the apartment type characteristics with the corresponding independent identification code of the data;

[0191] When the deviation data or identification error data exceeds the preset threshold, incremental learning iteration is triggered, or all data is aggregated at fixed times every day for batch incremental learning iteration.

[0192] During the iteration process, the dataset was classified and trained according to the skill level of construction personnel, the type of work process, and the characteristics of the apartment type. The parameters of the construction process image recognition model, the material category recognition and surplus calculation model, the material demand calculation model, and the violation behavior recognition model were optimized respectively. The parameters of the iterated models were matched and adapted according to the classification labels, the corresponding target construction areas, and the independent identity codes.

[0193] In practical applications, the specific steps are as follows:

[0194] C1. Iterative training: Collection and preprocessing steps of the full data source.

[0195] C11. Fixed-period full-volume data collection: The scheduling controller sets a fixed data collection period and synchronously collects three types of core iterative data sources. The first type is material supply and demand matching deviation data, which is the difference between the predicted demand of each material supply and demand matching judgment result and the actual material consumption data after the completion of the corresponding process, and calculates the deviation rate data. The second type is visual recognition model verification data, which is the automatic recognition results of the construction process image recognition model, material category recognition and surplus calculation model, compared with the standard results of manual review by project management personnel, and statistically analyzes the recognition error rate, false detection rate and false negative rate. The third type is violation identification model error data, which is the recognition results of the violation behavior identification model, compared with the results of manual review, and statistically analyzes the model's misjudgment and false negative data.

[0196] C12. Data Preprocessing: Standardize and preprocess the collected full data to remove outliers, missing values, and duplicate data. This is a common technique in the field of data processing and will not be described in detail in this embodiment. After preprocessing, the effective data is classified and stored in the local iterative training database of the scheduling controller to provide a standardized data source for subsequent iterative optimization.

[0197] C2. Steps for multi-dimensional association and labeling of iterative data:

[0198] C21. Retrieval of core related dimension data: For each piece of valid iterative data, based on the video stream frame-level identity identifier to which the data belongs and the independent identity code corresponding to the material supply and demand matching record, the corresponding dimension tag data is retrieved from the local database of the scheduling controller, including three types of core tags: construction personnel skill level information, process type of target construction area, and unit type characteristics.

[0199] C22. Data Tagging and Binding: The three types of tagged data are uniquely bound to the corresponding iterative data, and a standardized classification label is added to each data point. The label format is skill level-process type-household feature. After tagging, the dataset is classified and archived according to the label type to form a labeled iterative training dataset, ensuring that subsequent training can be accurately optimized according to different scenario dimensions.

[0200] C3. Execution steps of the dual-trigger mechanism for incremental learning iteration:

[0201] C31. Preset Threshold Setting and Instant Trigger Mechanism: Preset iterative trigger thresholds, where the preset threshold for material supply and demand matching deviation rate is 10%, and the preset threshold for recognition error rate between visual recognition model and violation recognition model is 5%. The scheduling controller monitors each newly collected valid data. When the deviation data or recognition error data of a single data exceeds the preset threshold, it immediately triggers incremental learning iteration, retrieves the data and nearly 100 sets of historical data with the same label, completes a single incremental learning, quickly corrects sudden errors in the model, and avoids continuous decline in accuracy.

[0202] C32. Fixed-period batch triggering mechanism: Set low-work periods as fixed-batch iteration windows, and summarize the full amount of labeled iterative training dataset for the day within the fixed-batch iteration window to trigger batch incremental learning iteration; adopt the well-known and mature fine-tuning training process in the field of incremental learning, without changing the basic backbone network of the model, only optimizing the fully connected layers and prediction head parameters of the model based on the newly added dataset, avoiding model overfitting, while not occupying the computing resources of the project's daytime work, and not affecting the normal execution of on-site scheduling and recognition tasks;

[0203] C4. Classification training and model parameter adaptation and optimization steps:

[0204] C41. Label-based Classification Training: During the iterative process, based on the labeled iterative training dataset, classification training is performed according to three dimensions: construction worker skill level, process type, and apartment type characteristics. For each label dataset, the parameters of the four core models are optimized, including: for the construction process image recognition model, material category recognition and surplus calculation model, and violation behavior recognition model, optimizing feature extraction weights and prediction thresholds; for the material demand calculation model, i.e., the multiple linear regression model, optimizing regression coefficients and loss prediction weights. During the classification training process, only the dataset corresponding to the label is used to complete the parameter optimization to ensure that the optimized model parameters can accurately adapt to the characteristics of the corresponding scenario.

[0205] C42. Model Parameter Matching, Adaptation and Deployment: The model parameters optimized after classification training are matched and adapted with the corresponding target construction area and independent identification code according to the corresponding classification label, and stored in the local database of the scheduling controller; when the scheduling controller processes video stream parsing, material demand calculation and violation identification tasks for the corresponding independent identification code and construction area, it automatically retrieves the matched and adapted model parameters to complete the inference calculation.

[0206] C43. Model Iteration Effect Verification and Archiving: After each iteration, the optimized model is verified for accuracy using a test set to ensure that the accuracy of the optimized model is better than that of the model before the iteration. After verification, the model parameters, iteration dataset, and accuracy improvement results after the iteration are archived and stored to form a full lifecycle ledger of model iteration, ensuring that the entire process is traceable and verifiable.

[0207] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for dynamic scheduling and optimization of intelligent construction resources based on industrial data analysis, applied to material resource scheduling in large-scale building indoor construction scenarios, characterized in that: include: Each construction worker involved in indoor construction is assigned a unique identification code. The unique identification code is then bound to the target construction area, preset construction procedures, and standard material requirement list corresponding to that construction worker, forming bound data. This bound data is then stored in the local database of the scheduling controller. Each construction worker is equipped with a monitor that is uniquely bound to their own unique identification code. The monitor's field of view is controlled to rotate synchronously with the construction worker's field of view, collect video data streams of the construction scene, and transmit the video stream data carrying the corresponding unique identification code to the scheduling controller. The scheduling controller parses and processes the received video data stream, identifies the completed process nodes within the current target construction area based on a pre-trained construction process image recognition model, and calculates the construction progress data. When the monitor detects that its field of view is directed toward the fixed material storage area within the target construction area, the remaining quantity calculation process is triggered. Based on the pre-trained material category identification model and remaining quantity calculation model, the remaining quantity data of each category of material in the fixed material storage area is extracted. The scheduling controller takes the standard material requirements list and construction progress data as input, calculates the theoretical material requirements required to complete the remaining construction procedures in the current target construction area, matches and verifies the theoretical material requirements with the remaining quantity data of the corresponding material categories, and generates a material supply and demand matching judgment result. Based on the material supply and demand matching determination result, if the current remaining material quantity is insufficient to support the completion of the remaining construction processes, the scheduling controller generates a material replenishment scheduling instruction for the corresponding category and quantity, and transfers the required materials to the fixed material storage area of ​​the target construction area through the material transfer unit; if it is identified that there are redundant materials of the same type in the target construction area that exceed the demand of its remaining processes, the scheduling controller generates a cross-regional material scheduling instruction, and transfers the redundant materials to the target construction area with insufficient remaining material quantity.

2. The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis according to claim 1, characterized in that, Each construction worker involved in indoor construction is assigned a unique identification code, which is then linked to the worker's target construction area, pre-defined construction procedures, and standard material requirements list. Specifically, this includes: Based on the BIM model of the target construction area, the schedule of the pre-set construction procedures, and the material consumption quota per unit procedure, calculate the standard material usage for a single procedure and the total usage for the entire procedure, and generate a standard material requirement list for the corresponding target construction area. The skill level information and safety training record information of the corresponding construction personnel are synchronously bound and stored with their unique identification codes.

3. The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis according to claim 1, characterized in that, The video data stream collected from the construction scene specifically includes: The monitor collects head posture data of construction workers, and the monitor is driven to adjust the shooting angle synchronously with the head rotation of the construction workers, so that the image collected by the monitor matches the field of vision of the construction workers. The monitor adds a unique identifier to the collected video stream data and transmits the identified video stream data to the scheduling controller via a wireless LAN.

4. The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis according to claim 1, characterized in that, The calculated construction progress data specifically includes: A pre-trained target detection model based on the YOLO deep learning framework is used as the image recognition model for construction procedures. The dataset of the pre-trained target detection model is constructed by collecting process node images and process completion label images corresponding to the construction procedure standards. The construction process image recognition model identifies the process completion nodes in the video stream data, counts the number of completed processes, calculates the progress completion rate of the completed processes in the current target construction area relative to the total processes, and simultaneously extracts the schedule plan of the remaining processes, which together serve as the construction progress data.

5. The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis according to claim 2, characterized in that, Extracting the remaining quantity data of each type of material in the fixed material storage area specifically includes: The individual volume and unit volume weight parameters of the corresponding material category are pre-stored in the local database of the scheduling controller; Identify the types of materials in the fixed material storage area and calculate the storage volume of the corresponding material type; By combining the individual volume and unit volume weight parameters of the corresponding material category pre-stored in the local database of the scheduling controller, the remaining quantity data of the material category is calculated.

6. The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis according to claim 1, characterized in that, The calculation of the theoretical material requirements for completing the remaining construction procedures in the current target construction area also includes a dynamic material loss correction step, which specifically includes: Using a multiple linear regression model in industrial data analysis, with historical material and reagent loss data, construction personnel skill levels, and construction environment parameters of the same type of construction process as input, the dynamic material loss correction coefficient for the corresponding process is calculated. The theoretical material demand is corrected by a dynamic material loss correction coefficient to obtain the corrected actual demand. The corrected actual demand is then matched and verified with the remaining quantity data to generate a material supply and demand matching result.

7. The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis according to claim 1, characterized in that, It also includes a step for monitoring construction violations during the construction process: The video stream data collected by the monitor is used to identify the compliance of construction workers' construction steps based on a pre-trained violation recognition model. When the compliance of construction steps is abnormal, a violation record is generated. Violation records are linked to the corresponding construction personnel's unique identification code and stored in the local database of the dispatch controller.

8. The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis according to claim 6, characterized in that, The dispatch controller generates material replenishment dispatch instructions for the corresponding category and quantity, specifically including: Based on the material supply and demand matching results, combined with the construction progress data corresponding to the independent identification code bound to the target construction area, the schedule plan of the remaining processes, and the skill level information of the construction personnel, the urgent priority of material shortages is divided. The system synchronously acquires inventory data of all material storage points in large-scale buildings, location and load status data of each material transfer unit, and on-site transfer path data generated based on the BIM model of the target construction area. It uses the highest priority of material shortage urgency, the lowest risk at the end of the process, the best transfer timeliness, and the lowest transfer cost as the multi-objective optimization function, and the corrected actual demand as the minimum loading constraint for a single transfer. It solves the optimal transfer path, transfer sequence, and loading allocation scheme through an improved multi-objective genetic algorithm in industrial data analysis, and generates corresponding material replenishment scheduling instructions based on the optimal scheme.

9. The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis according to claim 6, characterized in that, The specific steps involved in the dispatch controller generating cross-regional material dispatch instructions include: For all target construction areas, based on the construction progress data, remaining construction procedures, and dynamic material loss correction coefficient corresponding to the bound independent identity code, calculate the safety reserve of the same type of materials in each target construction area, and determine the part of the remaining material quantity that exceeds the safety reserve as schedulable redundant material. By using the global supply and demand balance optimization model in industrial data analysis, all schedulable redundant materials and material shortages are globally matched. The optimization objectives are: priority for materials from the same building and floor, priority for process sequence connection, shortest total transfer distance, and shortest total transfer period. The appearance integrity and specification matching degree of schedulable redundant materials are verified by video stream data collected by the monitor, and the globally optimal cross-regional material scheduling scheme is obtained. Based on the optimal solution, corresponding cross-regional material scheduling instructions are generated. At the same time, the scheduling material information is associated with the independent identification code bound to the target construction area, and the remaining data of the corresponding standard material demand list is updated.

10. The intelligent construction resource dynamic scheduling optimization method based on industrial data analysis according to claim 8, characterized in that, It also includes model self-optimization iteration steps, including: collecting deviation data between each material supply and demand matching judgment result and actual material consumption data, verification data of the recognition results of the construction process image recognition model and the material category recognition and surplus calculation model, and misjudgment and omission data of the violation recognition model; Associate the construction worker's skill level, the process type of the target construction area, and the apartment type characteristics with the corresponding independent identification code of the data; When the deviation data or identification error data exceeds the preset threshold, incremental learning iteration is triggered, or all data is aggregated at fixed times every day for batch incremental learning iteration. During the iteration process, the dataset was classified and trained according to the skill level of construction personnel, the type of work process, and the characteristics of the apartment type. The parameters of the construction process image recognition model, the material category recognition and surplus calculation model, the material demand calculation model, and the violation behavior recognition model were optimized respectively. The parameters of the iterated models were matched and adapted according to the classification labels, the corresponding target construction areas, and the independent identity codes.

Images