A Big Data-Based Intelligent Engineering Progress Supervision System and Method

Abstract

This invention discloses an intelligent engineering progress monitoring system and method based on big data, belonging to the field of big data analysis technology. The method includes the following steps: Step 1: Determine the number of engineering stages and projects by combining the BIM model; acquire historical engineering data and collect new engineering data in real time, and preprocess the data; Step 2: Construct the historical engineering progress into a uniform-size tensor, extract and aggregate features through masked convolution and cross-dimensional attention mechanisms to enhance effective data information; Step 3: Encode the real-time progress data of the new project and concatenate it with the feature tensor, construct a prediction model using an RNN, and train the prediction progress and completion status using a hybrid loss function; Step 4: Combine the predicted completion status and predicted progress of the new project to provide corresponding early warnings or feedback for projects that may be delayed or abnormal. This invention can effectively improve the situation of insufficient utilization of historical engineering data in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, specifically to an intelligent engineering progress monitoring system and method based on big data. Background Technology

[0002] As a core component of project management, project progress supervision aims to ensure on-time project delivery through dynamic monitoring and prediction. With the development of Building Information Modeling (BIM), the visualization and deconstruction of engineering structures through 3D digital modeling provides a precise spatial dimension for progress management. The widespread adoption of the Internet of Things (IoT) and smart terminals (such as drones and sensors) makes real-time data collection at construction sites possible, building a dynamic data flow covering the entire "design-construction-acceptance" cycle. Meanwhile, big data and artificial intelligence technologies (such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) provide algorithmic support for feature learning and time-series prediction of massive amounts of heterogeneous data.

[0003] Existing engineering progress supervision technologies have significant limitations in utilizing historical data, mainly in the following aspects: the phase divisions and project granularity of different projects vary significantly (e.g., some projects are divided into 5 phases, while others are divided into 8 phases). Traditional methods often store historical data in tabular form, lacking a unified structured modeling framework, resulting in incompatible data dimensions and making it difficult to directly use for comparative analysis of new projects. Existing models often use statistical analysis (e.g., mean, standard deviation) or shallow machine learning (e.g., random forest) to process historical data, which can only capture linear relationships or local features and cannot extract spatiotemporal cross-features. Temporal dependencies in historical data (e.g., the order of progress in each phase and resource competition among parallel projects) have not been effectively modeled. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent engineering progress monitoring system and method based on big data to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent project progress supervision method based on big data, the method comprising the following steps:

[0006] Step 1: Determine the number of engineering phases and projects based on the BIM model; acquire historical engineering data and collect new engineering data in real time, and preprocess the data;

[0007] Step 2: Construct historical project progress into a uniform-size tensor, and extract and aggregate features through mask convolution and cross-dimensional attention mechanisms to enhance effective data information;

[0008] Step 3: Encode the real-time progress data of the new project and concatenate it with the feature tensor. Use RNN to build a prediction model and train it with a hybrid loss function to predict the progress and completion status.

[0009] Step 4: Based on the predicted completion status and progress of the new project, issue corresponding warnings or feedback for projects that may be delayed or abnormal.

[0010] In step 1, engineering nodes are extracted from the BIM model, and the number of engineering stages n is automatically identified;

[0011] Based on the BIM model, hierarchical cluster analysis is performed on the construction tasks, and the number of projects m is determined according to the number of clusters in the clustering results.

[0012] Obtain historical engineering data from ERP systems or project management platforms, including the percentage of progress completed for each stage and the final project schedule achievement status.

[0013] For new engineering data, a real-time data acquisition network is built using IoT devices and smart terminals: the real-time progress of the project is obtained through manual data entry or by using drone aerial photography combined with BIM models.

[0014] Data cleaning includes deleting duplicate records, correcting erroneous data, and filling in missing values; and indicating whether the project was completed on schedule.

[0015] In step 2, the progress of the i-th historical project is represented by n. i ×m i The progress matrix; where n i m represents the number of the i-th historical engineering phase; i This represents the number of the i-th historical engineering project;

[0016] All historical projects are aligned to N. max ×M max Size, where N max M represents the maximum number of project phases across all historical projects. max This represents the maximum number of projects among all historical projects;

[0017] Fill in the mask values ​​for extra entries that exceed the actual stage or project, resulting in a tensor P∈R. K×Nmax×Mmax Where K represents the total number of historical projects;

[0018] Generate a corresponding Mask matrix based on the actual number of stages and projects for each historical project; for the i-th historical project, create N... max ×M max A Boolean Mask matrix, where n actually contains data. i ×m i Set the region to True and the rest of the regions to False;

[0019] When calculating convolution, a logical AND operation is performed between the convolution kernel and the corresponding position in the Mask matrix: when the corresponding position in the Mask matrix is ​​False, the contribution value of that position in the convolution operation is set to 0.

[0020] Feature tensor F∈R is obtained by extracting features through masked convolution. K×Nmax×Mmax×C Where C is the number of feature channels; the feature tensor is transformed to become F′∈R K×C×Nmax×Mmax ;

[0021] For N respectively max Dimensions and M max Dimensional calculation of attention weights: for N max The dimension is mapped from the feature tensor F′ to an intermediate dimension through a learnable linear layer, and then after passing through an activation function, it is mapped to N through another linear layer. max Dimension, to obtain N max The attention weight vector A of dimension N For M max Dimension, calculated to obtain M max The attention weight vector A of dimension M ;

[0022] Based on the calculated attention weights, the effective data locations of different cases are weighted and summed to obtain the aggregated feature tensor.

[0023] In step 3, the real-time progress data of the t-th stage of the project is encoded and concatenated with the feature tensor processed by mask convolution and cross-dimensional attention mechanism.

[0024] A recurrent neural network (RNN) is used to construct the time series prediction branch. The concatenated input tensor is sequentially fed into the RNN model, and the dependencies in the time series are captured through the iterative calculation of the hidden layer. In the last layer of the model, the hidden layer output is mapped to the required output dimension through a fully connected layer, which predicts the completion degree of each project in the subsequent stage and the probability of the overall project being completed on schedule.

[0025] For the task of predicting whether the overall project will be completed on schedule, a binary cross-entropy loss function is used: the predicted probability of the project being completed on schedule is represented by y', and the true label is represented by y, then the binary classification loss L b Represented as: L b =-y·log(y')-(1-y)·log(1-y');

[0026] For the task of predicting the project progress for each effective stage, the mean squared error loss function is used: the predicted progress is represented by p', the actual progress is p, and the regression loss is L. r Represented as: Lr =(1 / Q)·∑ q＝1 Q (p' q -p q ) 2 Where Q is the number of valid phase items, q∈{1,2,…,Q};

[0027] The weighted sum of the binary classification loss and the regression loss yields the total loss function L: L = α·L b +(1-α)·L r ; where α is a hyperparameter.

[0028] In step 4, when y' is not greater than the preset probability threshold β, it is considered that the project cannot be completed on schedule. The phased real-time progress is obtained, and the difference between the predicted progress and the actual progress is calculated, that is, the actual progress minus the predicted progress. When the difference is less than the preset threshold α1, the project corresponding to the progress is marked as a short-board project and feedback is given to the staff.

[0029] When y' is greater than the preset probability threshold β, the project is considered to be completed on schedule. Real-time progress is obtained in stages, and the difference between the predicted progress and the actual progress is calculated. When the difference is less than the preset threshold α2, an early warning is issued to the staff.

[0030] A big data-based intelligent project progress monitoring system, comprising a data preprocessing module, a feature processing module, a predictive modeling module, and an early warning feedback module;

[0031] The data preprocessing module is used to determine the number of engineering stages and projects by combining the BIM model; acquire historical engineering data and collect new engineering data in real time, and preprocess the data; the feature processing module is used to construct the historical engineering progress into a uniform-size tensor, and extract and aggregate features through mask convolution and cross-dimensional attention mechanisms to enhance effective data information; the prediction modeling module is used to encode the real-time progress data of new projects and concatenate it with the feature tensor, use RNN to build a prediction model, and train the prediction progress and completion status with a hybrid loss function; the early warning feedback module is used to combine the predicted completion status and predicted progress of new projects to provide corresponding early warnings or feedback for projects that may be delayed or abnormal.

[0032] The output of the data preprocessing module is connected to the input of the feature processing module; the output of the feature processing module is connected to the input of the prediction modeling module; and the output of the prediction modeling module is connected to the input of the early warning feedback module.

[0033] The data preprocessing module includes a BIM parsing unit and a data acquisition and cleaning unit;

[0034] The BIM parsing unit is used to automatically identify the number of engineering phases and projects based on the BIM model; the data acquisition and cleaning unit is used to acquire historical and real-time data from ERP or IoT devices, and perform data cleaning, missing value filling and annotation.

[0035] The output of the BIM parsing unit is connected to the input of the data acquisition and cleaning unit; the output of the data acquisition and cleaning unit is connected to the input of the feature processing module.

[0036] The feature processing module includes a tensor alignment unit, a mask convolution unit, and an attention aggregation unit;

[0037] The tensor alignment unit is used to uniformly align the historical project progress matrices of different sizes into tensors and fill them with Mask values; the mask convolution unit is used to filter invalid data and extract valid features through convolution operations with Mask matrices; the attention aggregation unit is used to calculate attention weights in the stage and project dimensions and aggregate features in a weighted manner.

[0038] The output of the tensor alignment unit is connected to the input of the mask convolution unit; the output of the mask convolution unit is connected to the input of the attention aggregation unit; and the output of the attention aggregation unit is connected to the input of the prediction modeling module.

[0039] The prediction modeling module includes a data encoding unit, an RNN prediction unit, and a hybrid training unit;

[0040] The data encoding unit is used to encode real-time progress data and concatenate it with historical feature tensors; the RNN prediction unit is used to establish a time-series prediction model through a recurrent neural network and output the predicted project progress and engineering completion status; the hybrid training unit is used to train the model by combining binary cross-entropy and mean squared error loss, and to set hyperparameters to balance the two types of tasks.

[0041] The output of the data encoding unit is connected to the input of the RNN prediction unit; the output of the RNN prediction unit is connected to the input of the hybrid training unit; and the output of the hybrid training unit is connected to the input of the early warning feedback module.

[0042] The early warning feedback module includes a progress comparison unit and a feedback early warning unit;

[0043] The progress comparison unit is used to calculate the difference between the predicted progress and the actual progress and identify abnormal items below the threshold; the feedback early warning unit is used to determine the project risk level based on the probability threshold and trigger early warning information or feedback on the weak points.

[0044] The output of the progress comparison unit is connected to the input of the feedback warning unit.

[0045] Compared with existing technologies, the beneficial effects of this invention are as follows: This application automatically identifies the number of stages and projects through BIM models, and combines hierarchical clustering analysis of construction tasks to achieve automated and standardized deconstruction of engineering structures, reduce human error, and improve the basic accuracy of data modeling; This application filters invalid areas through a Mask matrix, shields the influence of padding values ​​in convolution operations, and extracts features only from real data areas; At the same time, it introduces a cross-dimensional attention mechanism to dynamically focus on key stages and projects, strengthen the feature expression of effective data, and improve the model's adaptability to complex engineering structures; This application uses RNN to construct a prediction model, and captures the temporal correlation of project progress through hidden layer iterative calculations, more accurately predicting future progress trends and the probability of project completion. Attached Figure Description

[0046] Figure 1 This is a schematic diagram illustrating the steps of an intelligent engineering progress monitoring method based on big data according to the present invention.

[0047] Figure 2 This is a flowchart illustrating an intelligent engineering progress monitoring system based on big data, as described in this invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example: Figures 1-2 As shown, this invention provides a technical solution: an intelligent engineering progress monitoring method based on big data, which includes the following steps:

[0050] Step 1: Determine the number of engineering phases and projects based on the BIM model; acquire historical engineering data and collect new engineering data in real time, and preprocess the data;

[0051] Step 2: Construct historical project progress into a uniform-size tensor, and extract and aggregate features through mask convolution and cross-dimensional attention mechanisms to enhance effective data information;

[0052] Step 3: Encode the real-time progress data of the new project and concatenate it with the feature tensor. Use RNN to build a prediction model and train it with a hybrid loss function to predict the progress and completion status.

[0053] Step 4: Based on the predicted completion status and progress of the new project, issue corresponding warnings or feedback for projects that may be delayed or abnormal.

[0054] In step 1, engineering nodes are extracted from the BIM model, and the number of engineering stages n is automatically identified;

[0055] Based on the BIM model, hierarchical cluster analysis is performed on the construction tasks, and the number of projects m is determined according to the number of clusters in the clustering results.

[0056] Obtain historical engineering data from ERP systems or project management platforms, including the percentage of progress completed for each stage and the final project schedule achievement status.

[0057] For new engineering data, a real-time data acquisition network is built using IoT devices and smart terminals: the real-time progress of the project is obtained through manual data entry or by using drone aerial photography combined with BIM models.

[0058] Data cleaning includes deleting duplicate records, correcting erroneous data, and filling in missing values; and indicating whether the project was completed on schedule.

[0059] In step 2, the progress of the i-th historical project is represented by n. i ×m i The progress matrix; where n i m represents the number of the i-th historical engineering phase; i This represents the number of the i-th historical engineering project;

[0060] All historical projects are aligned to N. max ×M max Size, where N max M represents the maximum number of project phases across all historical projects. max This represents the maximum number of projects among all historical projects;

[0061] Fill in the mask values ​​for extra entries that exceed the actual stage or project, resulting in a tensor P∈R. K×Nmax×Mmax Where K represents the total number of historical projects;

[0062] Generate a corresponding Mask matrix based on the actual number of stages and projects for each historical project; for the i-th historical project, create N... max ×M max A Boolean Mask matrix, where n actually contains data. i ×m i Set the region to True and the rest of the regions to False;

[0063] When calculating convolution, a logical AND operation is performed between the convolution kernel and the corresponding position in the Mask matrix: when the corresponding position in the Mask matrix is ​​False, the contribution value of that position in the convolution operation is set to 0.

[0064] Feature tensor F∈R is obtained by extracting features through masked convolution. K×Nmax×Mmax×C Where C is the number of feature channels; the feature tensor is transformed to become F′∈R K×C×Nmax×Mmax ;

[0065] For N respectively max Dimensions and M max Dimensional calculation of attention weights: for N max The dimension is mapped from the feature tensor F′ to an intermediate dimension through a learnable linear layer, and then after passing through an activation function, it is mapped to N through another linear layer. max Dimension, to obtain N max The attention weight vector A of dimension N For M max Dimension, calculated to obtain M max The attention weight vector A of dimension M ;

[0066] Based on the calculated attention weights, the effective data locations of different cases are weighted and summed to obtain the aggregated feature tensor.

[0067] In step 3, the real-time progress data of the t-th stage of the project is encoded and concatenated with the feature tensor processed by mask convolution and cross-dimensional attention mechanism.

[0068] A recurrent neural network (RNN) is used to construct the time series prediction branch. The concatenated input tensor is sequentially fed into the RNN model, and the dependencies in the time series are captured through the iterative calculation of the hidden layer. In the last layer of the model, the hidden layer output is mapped to the required output dimension through a fully connected layer, which predicts the completion degree of each project in the subsequent stage and the probability of the overall project being completed on schedule.

[0069] For the task of predicting whether the overall project will be completed on schedule, a binary cross-entropy loss function is used: the predicted probability of the project being completed on schedule is represented by y', and the true label is represented by y, then the binary classification loss L b Represented as: L b =-y·log(y')-(1-y)·log(1-y');

[0070] For the task of predicting the project progress for each effective stage, the mean squared error loss function is used: the predicted progress is represented by p', the actual progress is p, and the regression loss is L. r Represented as: L r =(1 / Q)·∑ q＝1Q (p' q -p q ) 2 Where Q is the number of valid phase items, q∈{1,2,…,Q};

[0071] The weighted sum of the binary classification loss and the regression loss yields the total loss function L: L = α·L b +(1-α)·L r ; where α is a hyperparameter.

[0072] In step 4, when y' is not greater than the preset probability threshold β, it is considered that the project cannot be completed on schedule. The phased real-time progress is obtained, the difference between the predicted progress and the actual progress is calculated, and when the difference is less than the preset threshold α1, the project corresponding to the progress is marked as a short-board project and feedback is given to the staff.

[0073] When y' is greater than the preset probability threshold β, the project is considered to be completed on schedule. Real-time progress is obtained in stages, and the difference between the predicted progress and the actual progress is calculated. When the difference is less than the preset threshold α2, an early warning is issued to the staff.

[0074] A big data-based intelligent project progress monitoring system, comprising a data preprocessing module, a feature processing module, a predictive modeling module, and an early warning feedback module;

[0075] The data preprocessing module is used to determine the number of engineering stages and projects by combining the BIM model; acquire historical engineering data and collect new engineering data in real time, and preprocess the data; the feature processing module is used to construct the historical engineering progress into a uniform-size tensor, and extract and aggregate features through mask convolution and cross-dimensional attention mechanisms to enhance effective data information; the prediction modeling module is used to encode the real-time progress data of new projects and concatenate it with the feature tensor, use RNN to build a prediction model, and train the prediction progress and completion status with a hybrid loss function; the early warning feedback module is used to combine the predicted completion status and predicted progress of new projects to provide corresponding early warnings or feedback for projects that may be delayed or abnormal.

[0076] The output of the data preprocessing module is connected to the input of the feature processing module; the output of the feature processing module is connected to the input of the prediction modeling module; and the output of the prediction modeling module is connected to the input of the early warning feedback module.

[0077] The data preprocessing module includes a BIM parsing unit and a data acquisition and cleaning unit;

[0078] The BIM parsing unit is used to automatically identify the number of engineering phases and projects based on the BIM model; the data acquisition and cleaning unit is used to acquire historical and real-time data from ERP or IoT devices, and perform data cleaning, missing value filling and annotation.

[0079] The output of the BIM parsing unit is connected to the input of the data acquisition and cleaning unit; the output of the data acquisition and cleaning unit is connected to the input of the feature processing module.

[0080] The feature processing module includes a tensor alignment unit, a mask convolution unit, and an attention aggregation unit;

[0081] The tensor alignment unit is used to uniformly align the historical project progress matrices of different sizes into tensors and fill them with Mask values; the mask convolution unit is used to filter invalid data and extract valid features through convolution operations with Mask matrices; the attention aggregation unit is used to calculate attention weights in the stage and project dimensions and aggregate features in a weighted manner.

[0082] The output of the tensor alignment unit is connected to the input of the mask convolution unit; the output of the mask convolution unit is connected to the input of the attention aggregation unit; and the output of the attention aggregation unit is connected to the input of the prediction modeling module.

[0083] The prediction modeling module includes a data encoding unit, an RNN prediction unit, and a hybrid training unit;

[0084] The data encoding unit is used to encode real-time progress data and concatenate it with historical feature tensors; the RNN prediction unit is used to establish a time-series prediction model through a recurrent neural network and output the predicted project progress and engineering completion status; the hybrid training unit is used to train the model by combining binary cross-entropy and mean squared error loss, and to set hyperparameters to balance the two types of tasks.

[0085] The output of the data encoding unit is connected to the input of the RNN prediction unit; the output of the RNN prediction unit is connected to the input of the hybrid training unit; and the output of the hybrid training unit is connected to the input of the early warning feedback module.

[0086] The early warning feedback module includes a progress comparison unit and a feedback early warning unit;

[0087] The progress comparison unit is used to calculate the difference between the predicted progress and the actual progress and identify abnormal items below the threshold; the feedback early warning unit is used to determine the project risk level based on the probability threshold and trigger early warning information or feedback on the weak points.

[0088] The output of the progress comparison unit is connected to the input of the feedback warning unit.

[0089] In this embodiment, a large commercial complex construction project is used as an example. The project includes a two-story underground parking garage, a six-story above-ground commercial building, and supporting facilities.

[0090] Step 1: Data Acquisition and Preprocessing;

[0091] The BIM model was used to analyze the construction project of the commercial complex, and the number of project stages n=8 was automatically identified, including the foundation engineering, main structure construction, mechanical and electrical equipment installation, interior and exterior decoration engineering, etc.

[0092] By performing hierarchical cluster analysis on the construction tasks, the number of projects (m=20) is determined according to the number of clusters in the clustering results, such as foundation excavation projects, reinforced concrete pouring projects, and air conditioning system installation projects.

[0093] Historical engineering data from similar commercial building projects in the past are obtained from the enterprise's ERP system and project management platform, covering the percentage of progress completed for each stage and the final project schedule. For example, for a formwork installation project in the main structure construction stage of a historical project, the percentage of completion at different time points and whether the project was ultimately completed on schedule.

[0094] For new project data collection, IoT devices and smart terminals are deployed at the construction site to build a real-time data collection network. Construction personnel manually report daily project progress via a mobile app, while drone aerial photography combined with BIM models is used to obtain real-time progress data for each stage of the project. For example, drones conduct weekly aerial photography of the construction site, and image recognition technology is used to compare the data with the BIM model to determine the actual progress of each construction area.

[0095] The acquired data was cleaned, duplicate progress records were deleted, and erroneous data, such as incorrect completion percentages, were corrected. Missing values ​​were filled by averaging adjacent data. Finally, it was noted whether each phase of the commercial complex construction project was completed on schedule.

[0096] Step 2: Feature extraction and aggregation;

[0097] The progress of the i-th historical project is represented by a progress matrix of size ni×mi. All historical projects are aligned to a size of Nmax×Mmax. In this embodiment, the maximum number of project stages Nmax = 10 and the maximum number of projects Mmax = 25 among all historical projects. Excess entries exceeding the actual stages or projects are filled with Mask values ​​to obtain the tensor P.

[0098] A corresponding Mask matrix is ​​generated based on the actual number of stages and projects for each historical project. Taking one historical project as an example, if ni=7 and mi=18, a 10×25 Boolean Mask matrix is ​​created, setting the 7×18 area containing actual data to True and the remaining areas to False.

[0099] During convolution computation, a logical AND operation is performed between the convolution kernel and the corresponding position in the Mask matrix. When the corresponding position in the Mask matrix is ​​False, its contribution to the convolution operation is set to 0. After masked convolution, features are extracted to obtain a feature tensor F. The feature tensor is then transformed to F′.

[0100] Attention weights are calculated for both the Nmax and Mmax dimensions. For the Nmax dimension, the feature tensor F′ is mapped to an intermediate dimension through a learnable linear layer, then passed through an activation function, and finally mapped back to the Nmax dimension through another linear layer, resulting in the Nmax-dimensional attention weight vector AN. For the Mmax dimension, the Mmax-dimensional attention weight vector AM is calculated. Based on the calculated attention weights, a weighted sum is applied to the effective data locations of different cases to obtain the aggregated feature tensor.

[0101] Step 3: Prediction model construction and training;

[0102] The real-time progress data of phase t of the commercial complex construction project is encoded and concatenated with the feature tensor processed by masked convolution and cross-dimensional attention mechanisms. A recurrent neural network (RNN) is used to construct the time-series prediction branch. The concatenated input tensor is sequentially fed into the RNN model, and the dependencies in the time series are captured through iterative computation in the hidden layers. In the last layer of the model, a fully connected layer maps the output of the hidden layers to the required output dimension, predicting the completion rate of each project in subsequent phases and the probability of the overall project being completed on schedule.

[0103] For the task of predicting whether a project will be completed on schedule, a binary cross-entropy loss function is used. If the probability of a project being completed on schedule in a given prediction is y' = 0.6, and the true label is y = 1 (actually completed on schedule), then the binary classification loss L... b = -1×log(0.6)-(1-1)×log(1-0.6).

[0104] For the task of predicting the project progress for each effective stage, the mean squared error loss function is used. If the predicted project progress is p' = 0.7, the actual progress is p = 0.8, and the number of effective stages is Q = 10, then the regression loss L... r =(1 / 10)·∑ q＝1 10 (0.7-0.8)2 .

[0105] The weighted sum of the binary classification loss and the regression loss yields the total loss function L: L = α·L b +(1-α)·L r The hyperparameter α = 0.3. By continuously adjusting the model parameters and minimizing the total loss function, the model training is completed.

[0106] Step 4: Early Warning and Feedback;

[0107] A preset probability threshold β = 0.7 is used. When the predicted probability of on-time completion y' is not greater than 0.7, the project is considered unable to be completed on schedule. Real-time progress is acquired at each stage, and the difference between the predicted and actual progress is calculated. For example, if a renovation project has a predicted progress of 0.6 and an actual progress of 0.5, with a preset threshold α1 = 0, the difference is -0.1, which is less than the preset threshold. Therefore, the project corresponding to this progress is marked as a bottleneck project, and feedback is sent to the staff to remind them to take measures to accelerate the project's progress.

[0108] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for intelligent monitoring of project progress based on big data, characterized in that: The method includes the following steps: Step 1: Determine the number of engineering phases n and the number of projects m based on the BIM model; acquire historical engineering data and collect new engineering data in real time, and preprocess the data; Step 2: Construct historical project progress into a uniform-size tensor, and extract and aggregate features through mask convolution and cross-dimensional attention mechanisms to enhance effective data information; Step 3: Encode the real-time progress data of the new project and concatenate it with the feature tensor. Use RNN to build a prediction model, capture the dependencies in the time series through the iterative calculation of the hidden layer, and train the prediction progress and completion status with a hybrid loss function. Step 4: Based on the predicted completion status and progress of the new project, issue corresponding warnings or feedback for projects that may be delayed or abnormal. In step 2, the progress of the i-th historical project is represented by n. i ×m i The progress matrix; where n i m represents the number of the i-th historical engineering phase; i This represents the number of the i-th historical engineering project; All historical projects are aligned to N. max ×M max Size, where N max M represents the maximum number of project phases across all historical projects. max This represents the maximum number of projects among all historical projects; Fill in the mask values ​​for extra entries that exceed the actual stage or project, resulting in a tensor P∈R. K×Nmax×Mmax Where K represents the total number of historical projects; Generate a corresponding Mask matrix based on the actual number of stages and projects for each historical project; for the i-th historical project, create N... max ×M max A Boolean Mask matrix, where n actually contains data. i ×m i Set the region to True and the rest of the regions to False; When calculating convolution, a logical AND operation is performed between the convolution kernel and the corresponding position in the Mask matrix: when the corresponding position in the Mask matrix is ​​False, the contribution value of that position in the convolution operation is set to 0. Feature tensor F∈R is obtained by extracting features through masked convolution. K×Nmax×Mmax×C Where C is the number of feature channels; the feature tensor is transformed to become F′∈R K×C×Nmax×Mmax ; For N respectively max Dimensions and M max Dimensional calculation of attention weights: for N max The dimension is mapped from the feature tensor F′ to an intermediate dimension through a learnable linear layer, and then after passing through an activation function, it is mapped to N through another linear layer. max Dimension, to obtain N max The attention weight vector A of dimension N For M max Dimension, calculated to obtain M max The attention weight vector A of dimension M ; Based on the calculated attention weights, the effective data locations of different cases are weighted and summed to obtain the aggregated feature tensor.

2. The intelligent project progress supervision method based on big data according to claim 1, characterized in that: In step 1, engineering nodes are extracted from the BIM model, and the number of engineering stages n is automatically identified; Based on the BIM model, hierarchical cluster analysis is performed on the construction tasks, and the number of projects m is determined according to the number of clusters in the clustering results. Obtain historical engineering data from ERP systems or project management platforms, including the percentage of progress completed for each stage and the final project schedule achievement status. For new engineering data, a real-time data acquisition network is built using IoT devices and smart terminals: the real-time progress of the project is obtained through manual data entry or by using drone aerial photography combined with BIM models. Data cleaning includes deleting duplicate records, correcting erroneous data, and filling in missing values; and indicating whether the project was completed on schedule.

3. The intelligent project progress supervision method based on big data according to claim 2, characterized in that: In step 3, the real-time progress data of the t-th stage of the project is encoded and concatenated with the feature tensor processed by mask convolution and cross-dimensional attention mechanism. A recurrent neural network (RNN) is used to construct the time series prediction branch; The concatenated input tensors are sequentially fed into the RNN model, and the dependencies in the time series are captured through iterative computation in the hidden layers. In the last layer of the model, the output of the hidden layers is mapped to the required output dimension through a fully connected layer, which predicts the completion rate of each project in the subsequent stages and the probability of the overall project being completed on schedule. For the task of predicting whether the overall project will be completed on schedule, a binary cross-entropy loss function is used: the predicted probability of the project being completed on schedule is represented by y', and the true label is represented by y, then the binary classification loss L b Represented as: L b =-y·log(y')-(1-y)·log(1-y'); For the task of predicting the project progress for each effective stage, the mean squared error loss function is used: the predicted progress is represented by p', the actual progress is p, and the regression loss is L. r Represented as: Where Q is the number of valid phase items, q∈{1,2,…,Q}; The weighted sum of the binary classification loss and the regression loss yields the total loss function L: L = α·L b +(1-α)·L r ; where α is a hyperparameter.

4. The intelligent project progress supervision method based on big data according to claim 3, characterized in that: In step 4, when y' is not greater than the preset probability threshold β, it is considered that the project cannot be completed on schedule. The phased real-time progress is obtained, the difference between the predicted progress and the actual progress is calculated, and when the difference is less than the preset threshold α1, the project corresponding to the progress is marked as a short-board project and feedback is given to the staff. When y' is greater than the preset probability threshold β, the project is considered to be completed on schedule. Real-time progress is obtained in stages, and the difference between the predicted progress and the actual progress is calculated. When the difference is less than the preset threshold α2, an early warning is issued to the staff.

5. A big data-based intelligent project progress monitoring system, applied to the big data-based intelligent project progress monitoring method described in any one of claims 1-4, characterized in that: The system includes a data preprocessing module, a feature processing module, a predictive modeling module, and an early warning feedback module; The data preprocessing module is used to determine the number of engineering stages and projects by combining the BIM model; acquire historical engineering data and collect new engineering data in real time, and preprocess the data; the feature processing module is used to construct the historical engineering progress into a uniform-size tensor, and extract and aggregate features through mask convolution and cross-dimensional attention mechanisms to enhance effective data information; the prediction modeling module is used to encode the real-time progress data of new projects and concatenate it with the feature tensor, use RNN to build a prediction model, and train the prediction progress and completion status with a hybrid loss function; the early warning feedback module is used to combine the predicted completion status and predicted progress of new projects to provide corresponding early warnings or feedback for projects that may be delayed or abnormal. The output of the data preprocessing module is connected to the input of the feature processing module; the output of the feature processing module is connected to the input of the prediction modeling module; and the output of the prediction modeling module is connected to the input of the early warning feedback module.

6. The intelligent engineering progress monitoring system based on big data according to claim 5, characterized in that: The data preprocessing module includes a BIM parsing unit and a data acquisition and cleaning unit; The BIM parsing unit is used to automatically identify the number of engineering phases and projects based on the BIM model; the data acquisition and cleaning unit is used to acquire historical and real-time data from ERP or IoT devices, and perform data cleaning, missing value filling and annotation. The output of the BIM parsing unit is connected to the input of the data acquisition and cleaning unit; the output of the data acquisition and cleaning unit is connected to the input of the feature processing module.

7. The intelligent engineering progress monitoring system based on big data according to claim 6, characterized in that: The feature processing module includes a tensor alignment unit, a mask convolution unit, and an attention aggregation unit; The tensor alignment unit is used to uniformly align the historical project progress matrices of different sizes into tensors and fill them with Mask values; the mask convolution unit is used to filter invalid data and extract valid features through convolution operations with Mask matrices. The attention aggregation unit is used to calculate attention weights and weighted aggregate features at the stage and project dimensions; The output of the tensor alignment unit is connected to the input of the mask convolution unit; the output of the mask convolution unit is connected to the input of the attention aggregation unit; and the output of the attention aggregation unit is connected to the input of the prediction modeling module.

8. The intelligent engineering progress monitoring system based on big data according to claim 7, characterized in that: The prediction modeling module includes a data encoding unit, an RNN prediction unit, and a hybrid training unit; The data encoding unit is used to encode real-time progress data and concatenate it with historical feature tensors; the RNN prediction unit is used to establish a time-series prediction model through a recurrent neural network and output the predicted project progress and engineering completion status; the hybrid training unit is used to train the model by combining binary cross-entropy and mean squared error loss, and to set hyperparameters to balance the two types of tasks. The output of the data encoding unit is connected to the input of the RNN prediction unit; the output of the RNN prediction unit is connected to the input of the hybrid training unit; and the output of the hybrid training unit is connected to the input of the early warning feedback module.

9. The intelligent engineering progress monitoring system based on big data according to claim 8, characterized in that: The early warning feedback module includes a progress comparison unit and a feedback early warning unit; The progress comparison unit is used to calculate the difference between the predicted progress and the actual progress and identify abnormal items below the threshold; the feedback early warning unit is used to determine the project risk level based on the probability threshold and trigger early warning information or feedback on the weak points. The output of the progress comparison unit is connected to the input of the feedback warning unit.

Images