Engineering construction progress intelligent tracking system

Abstract

The invention provides an engineering construction progress intelligent tracking system, and belongs to the technical field of data processing. The system comprises a data fusion module for performing time alignment and feature fusion on multi-dimensional information data acquired by a data acquisition module to generate a fusion feature vector; the construction component association module judges whether the construction personnel fall into a construction unit area or not based on the positioning data of the construction personnel, determines an active area of the construction unit by combining the weight of the sensor data, and calculates the association degree of the personnel and the component subunits by utilizing the spatial distribution and the data weight of the sensor; determining a final associated component subunit in combination with the residence time; and the progress evaluation module is used for constructing a multi-base learner model, splicing a plurality of groups of progress prediction values output by the multi-base learner model with the fusion feature vector to form a secondary feature matrix, carrying out deep integration on the secondary feature matrix by utilizing a meta learner, capturing a comprehensive relationship of different base learners, and predicting to obtain an overall engineering construction progress state.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an intelligent tracking system for engineering construction progress. Background Art

[0002] In the actual construction process, in order to ensure the safety and reliability of the project, especially for engineering construction similar to water conservancy projects involving multiple fields such as civil engineering, electromechanical equipment, electrical, materials, etc., as the scale of engineering construction projects continues to expand and become more complex, traditional construction progress management methods, such as manual recording and experience judgment, can no longer effectively cope with the complex environment of multiple tasks, multiple construction units, and multiple types of components. It is necessary to combine intelligent management systems to strengthen construction management, strictly control construction quality and progress, and promptly discover and solve problems in the project to ensure the stable operation of water conservancy projects and lay a solid foundation for subsequent use and maintenance.

[0003] Considering that the existing construction progress tracking management usually relies on manual observation and regular reporting, it is difficult to reflect the real-time status of the construction site in a timely manner, which makes the plan adjustment and problem solving lagging, resulting in poor data real-time performance and information isolation. In addition, the current construction progress tracking management system is mostly a single data collection mode, which fails to achieve multi-dimensional fusion and comprehensive analysis of data, and cannot fully explore the potential association of multi-source data. Secondly, due to the dynamic and complex activities on the construction site, the relationship between the construction unit and the component sub-unit is difficult to accurately determine, making it difficult for the system to effectively and dynamically evaluate the current construction progress, and the adaptability and intelligence level of the system still need to be improved. In addition, most of the existing progress evaluations rely on a single model, which fails to effectively explore the complementary advantages of different algorithms, thereby limiting the prediction accuracy and reliability. These problems are intertwined, and there is an urgent need for an intelligent construction progress tracking system to achieve multi-source data fusion, component dynamic matching and accurate progress prediction. Summary of the invention

[0004] The main purpose of the present invention is to provide an intelligent tracking system for engineering construction progress, which aims to realize the real-time collection and integration of multi-dimensional information data, efficiently identify the relationship between construction units and component sub-units, accurately predict the overall engineering construction progress, and dynamically update the construction plan on this basis.

[0005] The embodiment of the present invention provides an intelligent tracking system for engineering construction progress, which includes the following modules.

[0006] The data acquisition module collects multi-dimensional information data of the construction site, obtains the BIM model of different component sub-units corresponding to each construction unit in the project and the associated construction plan information. The multi-dimensional information data includes sensor data and real-time positioning data of construction personnel.

[0007] The data fusion module performs time alignment and feature fusion on the collected multi-dimensional information data to generate a fused feature vector.

[0008] The construction component association module determines whether the construction personnel fall into the construction unit area based on the positioning data of the construction personnel, determines the active area of ​​the construction unit based on the weight of the sensor data, calculates the degree of association between the construction personnel and the component sub-unit using the spatial distribution and data weight of the sensor, and determines the final associated component sub-unit based on the residence time.

[0009] The progress assessment module constructs a multi-base learner model. Each base learner independently evaluates the progress status of the construction unit and the component sub-unit, generates the corresponding progress prediction value of the construction unit, and concatenates the multiple groups of progress prediction values ​​output by the multi-base learner model with the fusion feature vector to form a secondary feature matrix. The meta-learner is used to deeply integrate the secondary feature matrix to capture the comprehensive relationship between different base learners and predict the overall construction progress status of the project.

[0010] The progress tracking module, combined with the construction plan information, tracks and updates the construction plan after the current moment based on the predicted value of the project construction progress at the current moment.

[0011] Furthermore, in the data acquisition module, the location acquisition technology used to collect the real-time location data of the construction workers is a combination of one or more of ultra-wideband positioning, GPS and Bluetooth beacons.

[0012] The sensor components in the data acquisition module include:

[0013] Stress sensors are used to monitor the bearing capacity of key construction areas.

[0014] Temperature and humidity sensors are used to monitor the temperature and humidity in different areas of the construction site.

[0015] Vibration sensors are used to monitor the vibration intensity and frequency of construction equipment.

[0016] Magnetic field sensors are used to detect magnetic fields generated by construction equipment.

[0017] Furthermore, the positioning data and sensor data are aligned in the same time window, and the features extracted from the two are directly spliced ​​and fused to generate a fused data set. The formula for forming the fused data set at time t is: D t ={(P i ,s1,s2,…,s k )|i=1,2,…,n}, where P i =(x i ,y i ,z i ) is the position coordinate of the i-th construction worker, and the sensor data St ={s1,s2,…,s k}(s j represents the data of the jth sensor).

[0018] Normalize the fused data set and construct the fused feature vector F i =[x i ,y i ,z i ,s 1,i ,s 2,i ,…,s k,i ], where s j,i Represents the jth sensor data associated with the i-th construction worker position.

[0019] Furthermore, the mapping between the construction personnel position and the construction unit is realized based on the ray method, and the construction personnel position coordinates P are determined. p =(x p ,y p ,z p ) falls within the geometric polygon boundary of the construction unit, which can be expressed by the formula: If count is an odd number, then the point (x p ,y p ,z p ) is inside the polygon, where A[j] = {(x1,y1,z1),(x2,y2,z2),…,(x m ,y m ,z m )} represents the polygonal boundary set of the construction unit.

[0020] Furthermore, based on the DBSCAN algorithm and combined with the weight of the sensor data, a cluster analysis is performed on the fused feature vectors. Specifically, for the feature vector F corresponding to the i-th construction worker position point, i , and define its neighborhood as: Let the eigenvector F j With the current eigenvector F i The spatial distance of ||P i -P j ||2 satisfies the set distance threshold ε, and the feature vector F j The sum of the corresponding sensor weighted response values Exceeds the set threshold threshold, where ε is the distance threshold, ω k is the weight of the k-th sensor data, s k,j is the sampling value of the k-th sensor at the j-th construction worker position, threshold is the lower limit of the effective sensor response, and the sensor with the current point F is selected. iOther points F whose spatial distance is close and whose corresponding sensor responses are significant enough i .

[0021] The clustering results form a set of active areas of construction units: C = {Cluster1, Cluster2,…, Cluster M}, each cluster Cluster m (m=1,2,…,M) represents the activity range of a construction unit.

[0022] Furthermore, based on the clustering results of the construction unit, the affinity between each point in the construction unit and the component subunit is calculated, and the construction unit and the component subunit are matched. The calculation formula of the affinity is:

[0023] Affinity(c j )=∑ Fi∈Clusterm ω i *s j,i *exp(-d i,j ).

[0024] Among them, c j is the center point of component subunit j, ω i is the weight of the sensor data in the fusion feature vector, s j,i is the jth sensor at the construction worker position P i The response value, d i,j =||P i -c j ||2 is the Euclidean distance from the construction worker to the center of the component.

[0025] Considering the residence time of construction workers around the component subunit, the component subunit with the longest residence time is selected for dynamic optimization, and the final associated component subunit is determined, which is expressed as:

[0026] Furthermore, random forest, support vector machine and gradient boosting decision tree are integrated to construct a multi-base learner model, the logistic regression model is used as the meta-learner, and a hierarchical framework is adopted to integrate the base learner and the meta-learner.

[0027] The input of the multi-base learner model is the fused feature vector based on positioning data and sensor data. Random forest, support vector machine and gradient boosting decision tree respectively evaluate the progress status of construction units and component sub-units, capture different feature patterns, and obtain three sets of construction progress prediction values.

[0028] The three groups of progress prediction values ​​and the fusion feature vector are concatenated to obtain a secondary fusion feature matrix, which is then input into the logistic regression model of the meta-learner to learn the prediction mode of the base learner and output the overall construction progress prediction value.

[0029] Furthermore, based on the three base learners of random forest, support vector machine and gradient boosting decision tree, the secondary feature of the mth sample is X m meta =[y 1m ,y 2m ,y 3m ],y ik is the progress prediction value of the i-th base learner for the m-th sample, then the secondary features of N samples are combined to obtain the secondary feature matrix X meta .

[0030] The secondary feature matrix X meta It is vertically concatenated with the fused feature vector to obtain the secondary fused feature matrix.

[0031] Furthermore, a data set is constructed based on the historical fusion feature vectors, and the data set is divided into a training set and a test set according to a preset ratio, which are used for training and testing of each base learner respectively.

[0032] The K-fold cross-validation method is used to divide the primary training set. A sub-training set used for testing is combined with a test set to test each trained base learner. Each base learner is trained and tested K times in a cycle. The K training results and test results of all base learners are combined respectively. The combined training results and test results are concatenated with the primary training set and the primary test set in the historical fusion feature vector to obtain the secondary training set and the secondary test set.

[0033] The meta-learner is trained and tested using the secondary training set and the secondary test set, respectively.

[0034] Furthermore, before combining the K test results of each base learner, the average of the K test results is calculated, and the average value is concatenated with the first test set in the historical fusion feature vector to obtain a second test set.

[0035] Furthermore, the overlap ratio between each component sub-unit is calculated. When the overlap ratio exceeds a set marking threshold, it is marked as an overlapping area, and the overlapping area is divided between the overlapping component sub-units.

[0036] Further, the overlapping area is divided between overlapping component sub-units, including.

[0037] Combined with the time plan information of the overlapping component subunits in the construction plan, a higher time priority is assigned to the component subunit with an earlier time.

[0038] The overlapping component subunits have the same temporal priority, and the geometric priority is assigned according to the geometric characteristics of the component subunits.

[0039] The overlapping area is occupied by the component subunit with higher priority.

[0040] Furthermore, when the overlap rate exceeds a set fusion threshold, and the set fusion threshold is greater than a set marking threshold, the overlapping area is divided using the geometric priorities corresponding to the component subunits as weighting coefficients.

[0041] Compared with the prior art, the present application has the following beneficial effects.

[0042] The intelligent tracking system for project construction progress provided by the present invention is based on multi-dimensional data fusion, combined with active area identification of construction units, hierarchical prediction framework of multi-base learners and meta-learners, and priority division strategy of overlapping areas, to achieve intelligent tracking and dynamic adjustment of project construction progress. By integrating multiple innovative technologies such as multi-source data fusion, active area identification of construction units, dynamic matching of component sub-units, hierarchical prediction framework, etc., it effectively solves the problems of poor real-time performance, low data utilization, and insufficient prediction accuracy in traditional construction progress tracking methods, effectively improves the automation, accuracy and intelligence level of construction progress tracking, and provides practical technical support for the refined management of large and complex projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 A schematic diagram of the structure of an intelligent system for tracking project construction progress provided by an embodiment of the present invention.

[0045] Figure 2 A schematic diagram of the workflow of a construction component association module in an intelligent engineering construction progress tracking system provided in an embodiment of the present invention.

[0046] Figure 3 A schematic diagram of the training process of the hierarchical prediction framework of the multi-base learner and the meta-learner in the intelligent tracking system of the engineering construction progress provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0049] It should be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0050] In the actual construction process, in order to ensure the safety and reliability of the project, especially for engineering construction similar to water conservancy projects involving multiple fields such as civil engineering, electromechanical equipment, electrical, materials, etc., as the scale of engineering construction projects continues to expand and become more complex, traditional construction progress management methods, such as manual recording and experience judgment, can no longer effectively cope with the complex environment of multiple tasks, multiple construction units, and multiple types of components. It is necessary to combine intelligent management systems to strengthen construction management, strictly control construction quality and progress, and promptly discover and solve problems in the project to ensure the stable operation of water conservancy projects and lay a solid foundation for subsequent use and maintenance.

[0051] With the expansion of the scale and complexity of engineering construction, construction progress tracking management has gradually become a key factor affecting project duration, quality and cost. However, traditional progress tracking management faces many challenges. On the one hand, multi-source data at the construction site (such as construction personnel positioning data, sensor data and construction plan information) has not been fully utilized, resulting in poor data real-time performance and information isolation. On the other hand, due to the dynamic and complex activities at the construction site, the relationship between construction units and component sub-units is difficult to accurately determine. In addition, most existing progress assessment methods rely on a single model and fail to effectively explore the complementary advantages of different algorithms, thus limiting the prediction accuracy and reliability.

[0052] Please refer to Figure 1 , a comprehensive explanation of the intelligent tracking system of engineering construction progress is given, which includes the following modules.

[0053] The data acquisition module collects multi-dimensional information data of the construction site, obtains the BIM model of different component sub-units corresponding to each construction unit in the project and the associated construction plan information. The multi-dimensional information data includes sensor data and real-time positioning data of construction personnel.

[0054] The data fusion module performs time alignment and feature fusion on the collected multi-dimensional information data to generate a fused feature vector.

[0055] The construction component association module determines whether the construction personnel fall into the construction unit area based on the positioning data of the construction personnel, determines the active area of ​​the construction unit based on the weight of the sensor data, calculates the degree of association between the construction personnel and the component sub-unit using the spatial distribution and data weight of the sensor, and determines the final associated component sub-unit based on the residence time.

[0056] The progress assessment module constructs a multi-base learner model. Each base learner independently evaluates the progress status of the construction unit and the component sub-unit, generates the corresponding progress prediction value of the construction unit, and concatenates the multiple groups of progress prediction values ​​output by the multi-base learner model with the fusion feature vector to form a secondary feature matrix. The meta-learner is used to deeply integrate the secondary feature matrix to capture the comprehensive relationship between different base learners and predict the overall construction progress status of the project.

[0057] The progress tracking module, combined with the construction plan information, tracks and updates the construction plan after the current moment based on the predicted value of the project construction progress at the current moment.

[0058] In summary, the intelligent tracking system for construction progress is based on multi-dimensional data fusion, combined with the active area identification of construction units, multi-base learners and meta-learners, to achieve intelligent tracking and dynamic adjustment of construction progress. By real-time collection and fusion of multi-source data at the construction site, data utilization and analysis depth are significantly improved; the component matching model based on spatial geometric information and dynamic behavior data identifies the active area of ​​the construction unit and improves the accuracy of the association between the construction unit and the component subunit; through the hierarchical prediction framework, the feature capture capabilities of different learners are fully utilized, and the advantages of different learners are integrated to greatly improve the accuracy and robustness of progress prediction. The overall system has a high level of intelligence and automation, providing a new idea for the refined management of complex engineering projects.

[0059] As an optional solution of the present application, in the data acquisition module, a location acquisition technology formed by one or more combinations of ultra-wideband positioning, GPS and Bluetooth beacons is used to collect real-time positioning data of construction workers, wherein ultra-wideband positioning (UWB) is achieved by arranging multiple base stations and tag devices worn by construction workers at the construction site, and its positioning data accuracy can reach 10-30cm. GPS is applied to outdoor scenes, and combined with differential technology, it can provide centimeter-level accuracy. For complex areas (such as equipment-intensive areas), Bluetooth beacons can cooperate with accelerometers and inertial measurement units (IMUs) for auxiliary positioning. Taking the comprehensive use of the above three types of multi-source fusion positioning as an example, the above methods are combined, and Kalman filtering is used for multi-sensor data fusion to improve the stability of position data. Secondly, the sensor components in the data acquisition module include: stress sensors, which are used to monitor the bearing capacity of key construction parts, determine whether a specific area is under load, and assist in related construction activities; temperature and humidity sensors, which are used to monitor the temperature and humidity of different areas of the construction site; vibration sensors, which are used to monitor the vibration intensity and frequency of construction equipment, identify the use location of mechanical equipment and its impact range, and assist in determining the activity area of ​​the construction unit; magnetic field sensors, which are used to detect the magnetic field generated by construction equipment, identify the working area of ​​specific equipment and possible related construction activities. The location data of construction personnel related to the identification of the construction area is collected through positioning equipment, and a variety of sensors are arranged in key components and construction areas to collect sensor data related to the construction status of component subunits in real time, and the above data is stored in the local cache.

[0060] As an optional solution of this application, the positioning data and the sensor data are aligned in the same time window, and the time synchronization of a person's position and the sensor data is performed through a unified timestamp to ensure data accuracy, and the features extracted from the two are directly spliced ​​and fused to generate a fused data set. The formula for forming the fused data set at time t is: D t ={(P i ,s1,s2,…,s k )|i=1,2,…,n}, where P i =(x i ,y i ,z i ) is the position coordinate of the i-th construction worker, and the sensor data S t ={s1,s2,…,s k},s j represents the data of the jth sensor, and k is the number of sensors.

[0061] Perform feature space mapping of the fused data, normalize the fused data set, and construct the fused feature vector F i =[x i ,y i ,zi ,s 1,i ,s 2,i ,…,s k,i ], where s j,i Represents the jth sensor data associated with the i-th construction worker position and is mapped to the BIM model of the construction site based on the coordinates.

[0062] By integrating the real-time positioning data of construction workers with data from a variety of sensors, the real-time and integrity of construction site data can be significantly improved, providing multi-dimensional data support for subsequent construction unit identification and component sub-unit matching.

[0063] As an alternative to this application, see Figure 2 In the construction component association module, the mapping between the construction personnel position and the construction unit is realized based on the ray method to determine the construction personnel position coordinates P p =(x p ,y p ,z p ) falls within the geometric polygon boundary of the construction unit, which can be expressed by the formula: If count is an odd number, then the point (x p ,y p ,z p ) is inside the polygon, where A[j] = {(x1,y1,z1),(x2,y2,z2),…,(x m ,y m ,z m )} represents the polygonal boundary set of the construction unit.

[0064] Based on the DBSCAN clustering algorithm, combined with the weight of the sensor data, the fusion feature vector is clustered and analyzed. Specifically, for the feature vector F corresponding to the i-th construction worker position point i , and define its neighborhood as: Let the eigenvector F j With the current eigenvector F i The spatial distance of ||P i -P j ||2 satisfies the set distance threshold ε, and the feature vector F j The sum of the corresponding sensor weighted response values ​​∑ K k＝1 ω k* s k,j Exceeds the set threshold threshold, where ε is the distance threshold, ω k is the weight of the k-th sensor data, s k,jis the sampling value of the k-th sensor at the j-th construction worker position, threshold is the lower limit of the effective sensor response, and the sensor with the current point F is selected. i Other points F whose spatial distance is close and whose corresponding sensor responses are significant enough i .

[0065] The clustering results form a set of active areas of construction units: C = {Cluster1, Cluster2,…, Cluster M}, each cluster Cluster m (m=1,2,…,M) represents the activity range of a construction unit.

[0066] As a further preferred solution, by analyzing the intensity and distribution of construction activities in the construction unit, the construction personnel and sensor data are matched and associated with specific component subunits. Based on the clustering results of the construction unit, the affinity between each point in the construction unit and the component subunit is calculated, and the construction unit and the component subunit are matched. The calculation formula of affinity is: Affinity(c j )=∑ Fi∈Clusterm ω i *s j,i *exp(-d i,j ), where c j is the center point of component subunit j, ω i is the weight of the sensor data in the fusion feature vector, s j,i is the jth sensor at the construction worker position P i The response value, d i,j =||P i -c j ||2 is the Euclidean distance from the construction worker to the center of the component.

[0067] Considering the residence time of construction workers around the component subunit, the component subunit with the longest residence time is selected for dynamic optimization, and the final associated component subunit is determined, which is expressed as:

[0068] Based on the fused data, the clustering algorithm is used to identify the active areas of the construction unit. The affinity model between the construction personnel's position and the geometric center of the component sub-unit is used, and the final associated component sub-unit is dynamically determined in conjunction with the residence time. Finally, the distribution results of the construction personnel in the component sub-unit are output, providing a basis for intelligent tracking of the construction progress.

[0069] As an optional solution of the present application, since there may be partial overlap between different component sub-units, these overlapping areas need to be processed. When processing the overlapping areas of component sub-units, the overlapping parts of two construction components are first detected by calculating the overlap ratio (IoU) between each component sub-unit. The calculation formula of the overlap ratio is IoU = A intersection / A union , where A intersection is the overlapping area of ​​construction components, A union The total area of ​​the construction components is equal to the sum of the areas of each construction component minus the area of ​​their intersection. When the overlap rate exceeds the set marking threshold, it is marked as an overlapping area, and the overlapping area is divided between the overlapping component subunits. For example, if the marking threshold is set to 0.35, if the overlap rate exceeds 0.35, it is marked as an overlapping area, and the overlapping part with an overlap rate lower than 0.35 is marked with the segmentation result of the deep learning segmentation model.

[0070] For the division of overlapping areas among overlapping component sub-units, the component sub-units with higher priorities can occupy the overlapping areas in the following order: 1) Combined with the time plan information of the overlapping component sub-units in the construction plan, assign higher time priorities to the component sub-units with earlier times; 2) When the overlapping component sub-units have the same time priority, assign geometric priorities based on the geometric features of the component sub-units. The overlapping areas are occupied by component sub-units with higher priorities. It should be noted that the time priority is only for situations where the actual construction progress is inconsistent with the construction plan, while the geometric priority is a supplement to the basic situation, which can achieve comprehensive coverage of the division of overlapping areas. The geometric features can consider area, shape, etc. The area priority is to highlight the construction components with larger areas. area =A construct / A total , where A construct is the area of ​​the construction component, A total The area priorities of all overlapping construction components are calculated for the areas of all construction components, and they are sorted by the size of the area priority. The shape priority can be judged by reflecting the shape complexity by the aspect ratio of the construction component, Aspect Ratio (AR) = Height / Width. Components with more regular shapes may have higher priorities.

[0071] As a further preferred solution, if the overlapping area is large and cannot be simply divided by priority, a fusion strategy, such as weighted average or geometric transformation, can be used to reasonably allocate the area to multiple components. When the overlap rate exceeds the set fusion threshold, such as setting the fusion threshold to 0.65, if the overlap rate exceeds 0.65 and the set fusion threshold is greater than the set marking threshold, the geometric priority corresponding to the component subunit is used as the weighting coefficient to divide the overlapping area.

[0072] Before matching and associating component sub-units, a reasonable and flexible division strategy is implemented for the overlapping areas of the component sub-units to ensure the reasonable attribution of the overlapping areas, ensure the continuity and rationality of the construction progress tracking data, and effectively avoid construction plan deviations caused by improper handling of overlapping areas. This further improves the tracking of construction progress and the management of component sub-units, and ensures the accuracy of the segmentation results and the integrity of construction information.

[0073] As an optional solution of the present application, in the progress evaluation module, three different machine learning algorithms such as random forest, support vector machine and gradient boosting decision tree are integrated to construct a multi-base learner model, and the logistic regression model is used as the meta-learner, and the base learner and the meta-learner are integrated in a hierarchical framework. Among them, for the base learner, the random forest (RF) can handle nonlinear relationships and is suitable for multi-dimensional feature processing of sensor data. The support vector machine (SVM) is suitable for processing high-dimensional data and can capture the boundaries of progress features. The gradient boosting decision tree (GBDT) iteratively optimizes the prediction error to improve the accuracy of progress prediction; for the meta-learner, the easy-to-interpret logistic regression (LR) is selected, which can effectively integrate the prediction results output by multiple base learners.

[0074] The input of the multi-base learner model is the fused feature vector based on positioning data and sensor data. Random forest, support vector machine and gradient boosting decision tree respectively evaluate the progress status of construction units and component sub-units, capture different feature patterns, and obtain three sets of construction progress prediction values.

[0075] Taking the prediction results of the three base learners of random forest, support vector machine and gradient boosting decision tree as examples, the three groups of progress prediction values ​​are concatenated with the fusion feature vector to obtain the secondary fusion feature matrix. The secondary feature of the mth sample is X m meta =[y 1m ,y 2m ,y 3m ],y ik is the progress prediction value of the i-th base learner for the m-th sample, then the secondary features of N samples are combined to obtain the secondary feature matrix X meta , then the secondary feature matrix X meta The secondary fusion feature matrix is ​​vertically spliced ​​with the fusion feature vector. The secondary fusion feature matrix is ​​input into the logistic regression model of the meta-learner to learn the prediction mode of the base learner and output the prediction value of the overall construction progress.

[0076] As an optional solution of the present application, the overall training process of the hierarchical prediction framework of the multi-base learner and the meta-learner is reasonably designed, and the specific design is as follows.

[0077] A data set is constructed based on the historical fusion feature vector, and the data set is divided into a training set and a test set according to a preset ratio, which are used for training and testing of each base learner respectively; the K-fold cross-validation method is used to divide the training set, and a sub-training set used for testing is combined with the test set to test each trained base learner respectively, and each base learner is trained and tested K times in a cycle, and the K training results and test results of all base learners are combined respectively, and the combined training results and test results are spliced ​​with the training set and the test set in the historical fusion feature vector respectively to obtain a secondary training set and a secondary test set; the meta-learner is trained and tested using the secondary training set and the secondary test set respectively.

[0078] See also Figure 3 , use 10-fold cross validation to divide the primary training set into 10 equal parts, use 9 of the primary training subsets to train the base learner, and then use the remaining one as the primary test subset, use the primary test subset and the primary test set to test the trained base learner, train and test 10 times according to the above process, get the training results and test results of different base learners, combine the training results of all base learners, and concatenate them with the primary training set to get the secondary training set. Similarly, combine the test results of all base learners and concatenate them with the primary test set to get the secondary test set, use the secondary training set and the secondary test set to train and test the meta learner, and you can get a multi-base learner and meta learner that can be used to predict the overall project construction progress. The hierarchical prediction framework can be expressed in percentage.

[0079] In order to realize the learning and integration of the advantages of multiple machine learning algorithms and simplify the overall training process of the hierarchical prediction framework of multi-base learners and meta-learners, the K test results of each base learner are averaged before being combined, and the average value is concatenated with the primary test set in the historical fusion feature vector to obtain a secondary test set.

[0080] As an optional solution of the present application, in the progress tracking module, in combination with the construction plan, based on the predicted value of the construction progress at the current moment, the construction plan after the current moment is tracked and updated. Specifically: obtain the construction plan of the previous moment, convert it into the percentage of the construction plan progress based on the time plan information, and compare the predicted value of the construction progress percentage at the current moment with the percentage in the plan. If it is lower than the planned progress percentage, it means that the construction progress of the construction unit is delayed. If it is higher than the planned progress percentage, it means that the progress is ahead of schedule. The construction plan needs to be corrected and updated for both delayed and advanced progress. Taking the progress delay as an example, the percentage of delay can be converted into actual days or workload, and then the construction plan is adjusted later. The correction formula is: T_t n adjust=T_t n-1 plan+(T_t n-1 total*P_t n prediction), where T_t n adjust is the adjusted time, T_t n-1 plan is the planned construction period at the last moment, P_t n prediction is the predicted value of the construction progress percentage. For the case of progress delay, if the total planned construction period of a construction unit is 10 days and the current progress is 10% behind, P_t n prediction is taken as the percentage of delay P_t n If delay=0.1, the adjusted duration will be postponed by 1 day, that is, the adjusted time is T_t n-1 plan+1. For the situation where the progress is ahead of schedule, if the total planned duration of a construction unit is 20 days and the progress is ahead of schedule by 5% at the current moment, P_t n prediction is taken as the percentage of advance P_t n advance=-0.05), then the subsequent construction time can be advanced by 1 day, that is, the adjusted time is T_t n- 1plan–1 can advance the start time of subsequent construction steps, thereby making advance adjustments to subsequent tasks.

[0081] After correcting each construction unit by progress percentage, it is necessary to conduct a global assessment of the construction plan of the entire project. The critical path method (CPM) or Gantt chart in project management can be used to optimize the overall construction progress and reasonably allocate the construction period to ensure that the dependency of tasks will not cause global construction period conflicts due to the adjustment of individual units. For example, some tasks need to rely on the completion of previous tasks (such as the upper construction can only be carried out after the column construction is completed), so it is necessary to ensure that the critical path is not affected by the adjustment.

[0082] Based on the overall construction progress forecast output by the meta-learner, the progress status of the component subunits (normal, delayed or ahead of schedule) is automatically identified, and the subsequent construction tasks are dynamically adjusted in combination with the construction plan, which can effectively reduce the delay caused by the deviation between the actual construction progress and the original plan, while optimizing the construction efficiency and ensuring the flexibility and real-time nature of the construction plan. This real-time progress correction method based on actual data significantly improves the efficiency and accuracy of construction management, enhances the intelligent management level of the construction progress, and reduces the risk of project delays.

[0083] In summary, the intelligent tracking system for construction progress provided by the embodiment of the present invention is based on multi-dimensional data fusion, combined with the identification of active areas of construction units, the hierarchical prediction framework of multi-base learners and meta-learners, and the priority division strategy of overlapping areas, to achieve intelligent tracking and dynamic adjustment of construction progress. By real-time collection and fusion of spatial geometric information and dynamic behavior multi-source data of the construction site, the data utilization rate and analysis depth are significantly improved; the dynamic matching technology of construction units based on multi-source data fusion solves the problems of data isolation and insufficient matching accuracy in traditional methods, and improves the accuracy of the association between construction units and component sub-units; the priority division strategy for the overlapping problem of component sub-units effectively avoids the impact of the overlapping of component sub-units on the construction plan, and ensures the continuity and logic of the construction plan; through the hierarchical prediction framework, the advantages of different learners are integrated to greatly improve the accuracy and robustness of the progress prediction. The overall system has a high level of intelligence and automation, which provides a new idea for the refined management of complex engineering projects.

[0084] It should be understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.

[0085] The serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. An intelligent tracking system for construction progress, characterized in that: The system comprises: The data acquisition module collects multi-dimensional information data of the construction site, and obtains the BIM model of different component sub-units corresponding to each construction unit in the project and the associated construction plan information. The multi-dimensional information data includes sensor data and real-time positioning data of construction personnel; The data fusion module performs time alignment and feature fusion on the collected multi-dimensional information data to generate a fused feature vector; The construction component association module determines whether the construction personnel fall into the construction unit area based on the construction personnel positioning data, determines the active area of ​​the construction unit based on the weight of the sensor data, calculates the degree of association between the construction personnel and the component sub-units by using the spatial distribution and data weight of the sensors, and determines the final associated component sub-units based on the residence time; The progress evaluation module constructs a multi-base learner model. Each base learner independently evaluates the progress status of the construction unit and the component sub-unit, generates the corresponding progress prediction value of the construction unit, and splices the multiple groups of progress prediction values ​​output by the multi-base learner model with the fusion feature vector to form a secondary feature matrix. The meta-learner is used to deeply integrate the secondary feature matrix to capture the comprehensive relationship between different base learners and predict the overall construction progress status of the project. The progress tracking module, combined with the construction plan information, tracks and updates the construction plan after the current moment based on the predicted value of the project construction progress at the current moment.

2. The intelligent construction progress tracking system according to claim 1 is characterized in that: In the data acquisition module, the location acquisition technology used to collect real-time positioning data of construction personnel is one or more combinations of ultra-wideband positioning, GPS and Bluetooth beacons; the sensor components in the data acquisition module include: stress sensors, used to monitor the bearing capacity of key construction parts; temperature and humidity sensors, used to monitor the temperature and humidity of different areas of the construction site; vibration sensors, used to monitor the vibration intensity and frequency of construction equipment; magnetic field sensors, used to detect the magnetic field generated by construction equipment.

3. The intelligent construction progress tracking system according to claim 1 is characterized in that: Align the positioning data and sensor data in the same time window, and directly concatenate and fuse the features extracted from the two to generate a fused data set. Normalize the fused data set and construct the fused feature vector.

4. The intelligent construction progress tracking system according to claim 1 is characterized in that: The mapping between the construction worker position and the construction unit is realized based on the ray method, and it is judged whether the construction worker position coordinates fall within the geometric polygon boundary of the construction unit.

5. The intelligent construction progress tracking system according to claim 4 is characterized in that: Based on the DBSCAN algorithm and combined with the weight of sensor data, cluster analysis is performed on the fused feature vector.

6. The intelligent construction progress tracking system according to claim 5 is characterized in that: According to the clustering results of the construction unit, the affinity between each point in the construction unit and the component subunit is calculated, the construction unit and the component subunit are matched, and the residence time of the construction personnel around the component subunit is considered. The component subunit with the longest residence time is selected for dynamic optimization to determine the final associated component subunit.

7. The intelligent construction progress tracking system according to claim 1 is characterized in that: A multi-base learner model is constructed by integrating random forest, support vector machine and gradient boosting decision tree. The logistic regression model is used as the meta-learner, and a hierarchical framework is adopted to integrate the base learner and the meta-learner. The input of the multi-base learner model is a fused feature vector based on positioning data and sensor data. Random forest, support vector machine and gradient boosting decision tree respectively evaluate the progress status of construction units and component sub-units, capture different feature patterns, and obtain three groups of construction progress prediction values. The three groups of progress prediction values ​​are concatenated with the fused feature vector to obtain a secondary fused feature matrix, which is input into the logistic regression model of the meta-learner to learn the prediction mode of the base learner and output the prediction value of the overall project construction progress.

8. The intelligent construction progress tracking system according to claim 7 is characterized in that: Based on three base learners, random forest, support vector machine and gradient boosting decision tree, the secondary feature of the mth sample is X m meta =[y 1m ,y 2m ,y 3m ],y ik is the progress prediction value of the i-th base learner for the m-th sample, then the secondary features of N samples are combined to obtain the secondary feature matrix X meta ; The secondary feature matrix X meta It is vertically concatenated with the fused feature vector to obtain the secondary fused feature matrix.

9. The intelligent construction progress tracking system according to claim 7 is characterized in that: A data set is constructed based on the historical fusion feature vectors, and the data set is divided into a training set and a test set according to a preset ratio, which are used for training and testing of each base learner respectively; The K-fold cross-validation method is used to divide the primary training set, and a sub-training set used for testing is combined with a test set to test each trained base learner. Each base learner is trained and tested K times in a cycle, and the K-times training results and test results of all base learners are combined. The combined training results and test results are spliced ​​with the primary training set and the primary test set in the historical fusion feature vector to obtain the secondary training set and the secondary test set. The meta-learner is trained and tested using the secondary training set and the secondary test set, respectively.

10. The intelligent construction progress tracking system according to claim 1 or 6, characterized in that: Calculate the overlap rate between each component sub-unit. When the overlap rate exceeds the set marking threshold, it is marked as an overlapping area, and the overlapping area is divided between the overlapping component sub-units. Specifically include: Combined with the time plan information of the overlapping component subunits in the construction plan, the component subunits with earlier time are assigned time priority; when the overlapping component subunits have the same time priority, the geometric priority is assigned according to the geometric characteristics of the component subunits; the overlapping area is occupied by the component subunit with higher priority; when the overlap rate exceeds the set fusion threshold, and the set fusion threshold is greater than the set marking threshold, the overlapping area is divided using the geometric priority corresponding to the component subunit as the weighting coefficient.

Images