An intelligent management system for water conservancy project construction progress

Abstract

The present invention relates to the technical field of water conservancy projects, and in particular to an intelligent management system for the construction progress of water conservancy projects. By acquiring geometric and semantic information, three-dimensional geological parameter distribution maps, and mechanical posture data from a BIM model, the acquired data is preliminarily processed, the geometric and semantic information in multimodal water conservancy construction data is segmented, and the time series data is mapped to a process state space to obtain cross-modal semantic alignment data. The cross-modal semantic alignment data is then represented by a spatiotemporal-physical joint embedding. A recognition model is established based on a PINN physical information neural network. A partial differential equation residual term is introduced into the model's loss function to obtain a construction progress deviation recognition model. The characteristic water conservancy construction data is input into the construction progress deviation recognition model for recognition, thereby obtaining a construction progress result. The system can accurately identify construction progress deviations and provide a reliable decision-making basis for water conservancy project construction management.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy projects, and in particular to an intelligent management system for the construction progress of water conservancy projects. Background Art

[0002] Current water conservancy project construction progress management faces numerous pressing challenges, severely hindering the efficient progress of projects. Planning currently relies primarily on manual experience. Construction managers manually compile construction schedules based on experience gained from previous projects, combined with design drawings and contract deadlines. However, the construction environment of water conservancy projects is complex and dynamic, with a constant stream of dynamic factors, such as sudden changes in geological conditions, frequent meteorological disasters, and interference from the surrounding environment. Manually formulated plans lack scientific quantitative analysis and dynamic adjustment mechanisms, making it difficult to accurately predict the impact of these complex conditions on the construction progress. When emergencies arise, plans often cannot be optimized in a timely manner, significantly increasing the risk of delays. Traditional methods, such as manual on-site inspections and paper-based report recording, are subject to significant lags. Construction progress information, from on-site collection to summary analysis, requires multiple steps, which is time-consuming and makes it difficult for management to obtain accurate, real-time data. When construction incidents occur, immediate response and decision-making are delayed, further negatively impacting project progress. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and to design an intelligent management system for the construction progress of water conservancy projects.

[0004] To achieve the above object, the technical solution of the present invention is that, further, in the above-mentioned intelligent management system for the construction progress of a water conservancy project, the intelligent management system for the construction progress of a water conservancy project includes the following modules:

[0005] The multimodal data acquisition module is used to obtain geometric and semantic information in the BIM model, obtain three-dimensional geological parameter distribution maps through geological radar equipment, and use sensors to obtain mechanical posture data of water conservancy construction sites. The acquired data is preliminarily processed to obtain multimodal water conservancy construction data;

[0006] A data feature extraction module is used to segment the geometric and semantic information in the multimodal water conservancy construction data, map the time series data to the process state space, obtain cross-modal semantic alignment data, and perform a spatiotemporal-physical joint embedding representation on the cross-modal semantic alignment data to obtain characteristic water conservancy construction data;

[0007] A recognition model building module is used to build a recognition model based on the PINN physical information neural network. A Transformer module is added to the recognition model to capture long-distance temporal dependencies, and a partial differential equation residual term is introduced into the model's loss function to obtain a PINN-Transformer construction progress deviation recognition model.

[0008] The construction progress identification module is used to input the characteristic water conservancy construction data into the PINN-Transformer construction progress deviation identification model for identification, and use the MAML algorithm to extract common feature representations in the training process to obtain construction progress results.

[0009] Furthermore, in the above-mentioned intelligent management system for water conservancy project construction progress, the multimodal data acquisition module includes the following submodules:

[0010] The BIM model information extraction submodule is used to use the Revit API secondary development tool to export the geometric parameters of the components, including at least three-dimensional coordinates, volume and surface area, and topological relationships; establish a water conservancy project ontology semantic library, including at least function, material and process categories, and use the NLP parser to extract the process parameters in the semantic library to obtain geometric and semantic information;

[0011] The geological radar data acquisition submodule is used to collect real-time scanning information of the geological radar equipment and convert the electromagnetic wave reflection delay and amplitude data in the real-time scanning information into a three-dimensional geological parameter distribution map using a conjugate gradient inversion algorithm;

[0012] The mechanical posture data acquisition submodule is used to obtain real-time data from excavators, tower cranes, and pump trucks. It includes at least a 9-axis IMU, GNSS-RTK, laser rangefinder, stress sensor, flow meter, pressure sensor, mechanical vibration frequency, and current load to obtain mechanical posture data.

[0013] The data preprocessing submodule is used to use the PTP protocol to time-align the geometric and semantic information, the three-dimensional geological parameter distribution map and the mechanical posture data, delete outliers from the aligned data, and obtain multimodal water conservancy construction data.

[0014] Furthermore, in the above-mentioned intelligent management system for water conservancy project construction progress, the data feature extraction module includes the following submodules:

[0015] A semantic segmentation submodule is used to segment the geometric and semantic information in the multimodal water conservancy construction data using a graph attention segmentation network, define a graph structure, where the vertex set is the BIM component and the edge set is the physical connection or process dependency of the component, and obtain segmented data;

[0016] A state mapping submodule, configured to map the mechanical vibration frequency and current load in the multimodal water conservancy construction data to a process state space to obtain cross-modal semantically aligned data;

[0017] A sample generation submodule is used to generate digital twins of the stress and seepage fields in the construction area based on finite element analysis, encode the physical field simulation results in the cross-modal semantic alignment data into high-dimensional feature vectors; establish positive and negative sample pairs, where the positive sample is the sensor data and physical field feature vector at the same time and space point, and the negative sample is a random combination of different construction stages or spatial locations;

[0018] The embedding representation submodule is used to train the contrastive learning model to maximize the similarity of positive samples and minimize the similarity of negative samples, and to perform a spatiotemporal-physical joint embedding representation on the cross-modal semantic alignment data to obtain characteristic water conservancy construction data.

[0019] Furthermore, in the above-mentioned intelligent management system for water conservancy project construction progress, the recognition model building module includes the following submodules:

[0020] Establish a submodule to build an identification model based on the PINN physical information neural network and define the partial differential equation of construction progress evolution:

[0021] ;

[0022] in, Indicates spatial location Department The progress completion rate at the moment, Represents a resource-driven item, represents the process coupling diffusion term, represents the environmental loss term, Indicates the resource input intensity, including at least machine shifts and man-hours; Represents environmental disturbance factors, including at least rainfall and temperature deviation; , and represents the learnable physical parameters, represents the Laplace operator term, which characterizes the process coupling diffusion effect of the construction progress in space;

[0023] Add submodules to build a dual-channel input architecture for the recognition model, including physical and process channels, and add a Transformer module to the recognition model to capture long-range temporal dependencies;

[0024] Define attention weights and physical mask matrices :

[0025] ;

[0026] in, It controls the decay rate of spatiotemporal correlation, and determines how fast the correlation decreases as the time and space distance increase. The larger the value, the faster the decay; and Represent two different moments, used to calculate the time difference, Represents two moments and The absolute value of the time difference between them is used to measure the interval in the time dimension; and Represent two different spatial locations respectively; Represents two spatial locations and The distance between them is used to measure the interval in the spatial dimension.

[0027] Furthermore, in the above-mentioned intelligent management system for water conservancy project construction progress, the recognition model building module further includes the following submodules:

[0028] Introducing a submodule for introducing partial differential equation residual terms into the loss function of the model, including at least data fitting terms, physical residual terms, and process constraint terms;

[0029] The total loss function is:

[0030] ;

[0031] in, Represents the total loss function of the recognition model, Represents the data fitting term, which characterizes the difference between the model prediction results and the actual observed data; represents the physical residual term, corresponding to the residual of the partial differential equation; Represents process constraints, which are constraints imposed by construction technology on progress; , and Represents the adaptive weight parameter, which is used to adjust the relative importance of the data fitting term, physical residual term, and process constraint term in the total loss function;

[0032] The adjustment submodule is used to adjust the adaptive weights to obtain the PINN-Transformer construction progress deviation identification model.

[0033] Furthermore, in the above-mentioned intelligent management system for water conservancy project construction progress, the construction progress identification module includes the following units:

[0034] A data input unit, configured to input the characteristic water conservancy construction data into the PINN-Transformer construction progress deviation identification model for identification and load model parameters;

[0035] A path processing unit is used to perform dual-path processing in the diagnosis layer of the identification model, wherein the physical path is based on the construction dynamics equation for mechanism deduction, and the data path captures long-term dependencies through Transformer;

[0036] The progress identification unit is used to generate multi-dimensional identification results, including at least the current progress deviation index, key bottleneck process location and potential risk event warning.

[0037] Furthermore, in the above-mentioned intelligent management system for water conservancy project construction progress, the construction progress identification module further includes the following subunits:

[0038] The meta-learner unit is used to extract common feature representations during training using the MAML algorithm, load the basic feature encoder from the cross-project meta-knowledge base, select the first three days of project data as the support set, freeze the physical constraint layer parameters, adjust the feature extraction layer, and use second-order gradient optimization to accelerate convergence;

[0039] The teacher-student architecture unit is used to construct the teacher-student model architecture, where the teacher model is a complete PINN-Transformer architecture and the student model is a lightweight diagnostic network.

[0040] Furthermore, in the method for implementing the above-mentioned intelligent management system for the construction progress of a water conservancy project, the method includes the following steps:

[0041] Acquire geometric and semantic information from the BIM model, obtain three-dimensional geological parameter distribution maps through geological radar equipment, use sensors to obtain mechanical posture data of the water conservancy construction site, and perform preliminary processing on the acquired data to obtain multimodal water conservancy construction data;

[0042] Segmenting the geometric and semantic information in the multimodal water conservancy construction data, mapping the time series data to the process state space to obtain cross-modal semantic alignment data, and performing a spatiotemporal-physical joint embedding representation on the cross-modal semantic alignment data to obtain characteristic water conservancy construction data;

[0043] A recognition model is established based on the PINN physical information neural network. A Transformer module is added to the recognition model to capture long-range temporal dependencies. A partial differential equation residual term is introduced into the model's loss function to obtain the PINN-Transformer construction progress deviation recognition model.

[0044] The characteristic water conservancy construction data is input into the PINN-Transformer construction progress deviation identification model for identification, and the MAML algorithm is used to extract common feature representations during the training process to obtain the construction progress results.

[0045] Furthermore, a method for implementing the above-mentioned intelligent management system for water conservancy project construction progress is characterized in that the method includes the following steps:

[0046] Based on the PINN physical information neural network, an identification model is established to define the partial differential equation of construction progress evolution;

[0047] A dual-channel input architecture of the recognition model is constructed, including a physical channel and a process channel, and a Transformer module is added to the recognition model to capture long-distance temporal dependencies.

[0048] Furthermore, a method for implementing the above-mentioned intelligent management system for water conservancy project construction progress is characterized in that the method includes the following steps:

[0049] Introducing partial differential equation residual terms into the loss function of the model, including at least data fitting terms, physical residual terms, and process constraint terms;

[0050] The adaptive weights are adjusted to obtain the PINN-Transformer construction progress deviation identification model.

[0051] Its beneficial effect is that by integrating BIM models, geological radar and sensor data, it realizes the comprehensive collection of multi-dimensional information of water conservancy construction sites. The segmentation of multimodal data, the mapping of time series data to process state space, and the joint spatiotemporal-physical embedding representation effectively solve the semantic alignment and feature fusion problems of cross-modal data. The PINN-Transformer construction progress deviation identification model has both physical information constraints and the ability to capture long-term time series dependencies. PINN introduces the residual term of the partial differential equation to integrate the physical laws of water conservancy construction into model training to ensure that the recognition results are consistent with the actual engineering physical characteristics; the Transformer module effectively captures the dependencies in long-distance time series data and enhances the model's ability to analyze the complex dynamic changes in the construction progress. Combining the MAML algorithm to extract common feature representations not only improves the efficiency of model training, but also enhances its generalization ability, and can accurately identify construction progress deviations, providing a scientific and reliable decision-making basis for water conservancy project construction management. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0053] Figure 1 This is a schematic diagram of a first embodiment of an intelligent management system for water conservancy project construction progress according to an embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of a second embodiment of an intelligent management system for water conservancy project construction progress according to an embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram of a third embodiment of an intelligent management system for water conservancy project construction progress in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0057] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", and "" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0058] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 As shown, an intelligent management system for the construction progress of a water conservancy project includes the following modules:

[0059] 101. Multimodal data acquisition module, used to obtain geometric and semantic information in the BIM model, obtain three-dimensional geological parameter distribution maps through geological radar equipment, obtain mechanical posture data of water conservancy construction sites using sensors, and perform preliminary processing on the obtained data to obtain multimodal water conservancy construction data;

[0060] Specifically, this embodiment also includes a BIM model information extraction submodule for using the Revit API secondary development tool to export the geometric parameters of the components, including at least three-dimensional coordinates, volume and surface area, and topological relationships; establishing a water conservancy project ontology semantic library, including at least function classes, material classes, and process classes, and using an NLP parser to extract the process parameters in the semantic library to obtain geometric and semantic information;

[0061] The geological radar data acquisition submodule is used to collect real-time scanning information of the geological radar equipment and convert the electromagnetic wave reflection delay and amplitude data in the real-time scanning information into a three-dimensional geological parameter distribution map using the conjugate gradient inversion algorithm;

[0062] The mechanical posture data acquisition submodule is used to obtain real-time data from excavators, tower cranes, and pump trucks. It includes at least a 9-axis IMU, GNSS-RTK, laser rangefinder, stress sensor, flow meter, pressure sensor, mechanical vibration frequency, and current load to obtain mechanical posture data.

[0063] The data preprocessing submodule is used to use the PTP protocol to time-align geometric and semantic information, three-dimensional geological parameter distribution maps, and mechanical posture data, delete outliers from the aligned data, and obtain multimodal water conservancy construction data.

[0064] 1. Deep extraction of BIM model information:

[0065] Geometric information extraction uses Revit API secondary development tools to batch export component geometric parameters, 3D coordinates (X, Y, Z); volume / surface area (concrete volume, formwork contact area); and topological relationships (component connectivity matrix) using the Octree space partitioning algorithm to achieve fast spatial retrieval of large scene models.

[0066] Extract semantic information, build a water conservancy project ontology library, and define a semantic tagging system: functional category (water retaining structure / diversion facility); material category (C30 concrete / HRB400 steel bar); process category (layered pouring / curtain grouting). Develop an NLP parser to automatically extract process parameters from BIM attributes: pouring temperature control range (e.g., 14-28°C); strength threshold for formwork removal (e.g., 75% of design strength);

[0067] The model is lightweight and LOD (Level of Detail) graded optimization is implemented: LOD3: complete geometric details are retained in key construction areas, LOD1: non-critical areas are simplified to bounding box representation; the Draco compression algorithm is used to achieve a triangle face data compression ratio of ≥8:1.

[0068] 2. Geological radar data acquisition and modeling:

[0069] The equipment deployment plan uses an array-type geological radar vehicle with the following configurations: transmission frequency: 100MHz (taking into account both penetration depth and resolution); scanning spacing: 0.5m (longitudinally arranged along the construction axis); detection depth: 30m (meeting the requirements for dam foundation detection);

[0070] Geological inversion modeling, implementing the improved conjugate gradient inversion algorithm: Input: electromagnetic wave reflection delay and amplitude data Output: three-dimensional dielectric constant distribution map, establishing geological parameter mapping relationship;

[0071] Geology-BIM integration, development of a non-uniform grid interpolation algorithm: aligning the geological model grid (1m×1m×0.5m) with the BIM model grid (0.2m accuracy);

[0072] Generate geologically enhanced BIM component properties: foundation excavation rock mass grade (Class I-V) and recommended support scheme (anchor length / spacing).

[0073] 3. Mechanical posture data perception:

[0074] The sensor networking solution installs multi-source sensors on key construction machinery: Excavators: 9-axis IMU (attitude angle) + GNSS-RTK (planar positioning ±1cm); tower cranes: laser rangefinder (hook position) + strain sensor (lifting capacity); pump trucks: flow meter (concrete delivery volume) + pressure sensor (pipeline pressure);

[0075] Data is aligned in time and space to establish a unified time and space benchmark: Time synchronization: PTP protocol is used to achieve μs-level time alignment; Spatial coordinates: the local mechanical coordinate system is converted to the engineering CGCS2000 coordinate system.

[0076] 4. Multimodal data preprocessing:

[0077] Standardize heterogeneous data formats and define unified data interface specifications: Geometric data: CityGML standard extended format; Sensor data: Apache Parquet column storage; Geological data: GeoTIFF raster encoding; Data quality enhancement; Implementation of abnormal data cleaning process: Eliminate outliers based on the 3σ criterion; Use KNN algorithm to fill in missing data; Apply wavelet threshold denoising to vibration noise.

[0078] 102. Data feature extraction module, used to segment the geometric and semantic information in multimodal water conservancy construction data, map the time series data to the process state space, obtain cross-modal semantic alignment data, perform spatiotemporal-physical joint embedding representation on the cross-modal semantic alignment data, and obtain characteristic water conservancy construction data;

[0079] Specifically, this embodiment also includes a semantic segmentation submodule for segmenting the geometric and semantic information in the multimodal water conservancy construction data using a graph attention segmentation network, defining a graph structure in which the vertex set is the BIM component and the edge set is the physical connection or process dependency of the component, thereby obtaining segmented data;

[0080] The state mapping submodule is used to map the mechanical vibration frequency and current load in the multimodal water conservancy construction data into the process state space to obtain cross-modal semantically aligned data;

[0081] The sample generation submodule is used to generate digital twins of the stress and seepage fields in the construction area based on finite element analysis. It encodes the physical field simulation results in the cross-modal semantic alignment data into high-dimensional feature vectors. It also establishes positive and negative sample pairs, where the positive sample is the sensor data and physical field feature vector at the same time and space point, and the negative sample is a random combination of different construction stages or spatial locations.

[0082] The embedding representation submodule is used to train the contrastive learning model to maximize the similarity of positive samples and minimize the similarity of negative samples, and to perform a spatiotemporal-physical joint embedding representation on the cross-modal semantic alignment data to obtain characteristic water conservancy construction data.

[0083] 103. Identification model building module, used to build an identification model based on the PINN physical information neural network. The Transformer module is added to the identification model to capture long-distance temporal dependencies. The residual term of the partial differential equation is introduced into the loss function of the model to obtain the PINN-Transformer construction progress deviation identification model.

[0084] Specifically, this embodiment also includes establishing a submodule for establishing a recognition model based on the PINN physical information neural network and defining the partial differential equation of the construction progress evolution:

[0085] ;

[0086] in, Indicates spatial location Department The progress completion rate at the moment, Represents a resource-driven item, represents the process coupling diffusion term, represents the environmental loss term, Indicates the resource input intensity, including at least machine shifts and man-hours; Represents environmental disturbance factors, including at least rainfall and temperature deviation; , and represents the learnable physical parameters, represents the Laplace operator term, which characterizes the process coupling diffusion effect of the construction progress in space;

[0087] Add submodules to build a dual-channel input architecture for the recognition model, including physical and process channels. Add a Transformer module to the recognition model to capture long-range temporal dependencies.

[0088] Define attention weights and physical mask matrices :

[0089] ;

[0090] in, It controls the decay rate of spatiotemporal correlation, and determines how fast the correlation decreases as the time and space distance increase. The larger the value, the faster the decay; and Represent two different moments, used to calculate the time difference, Represents two moments and The absolute value of the time difference between them is used to measure the interval in the time dimension; and Represent two different spatial locations respectively; Represents two spatial locations and The distance between them is used to measure the interval in the spatial dimension.

[0091] Introducing a submodule for introducing partial differential equation residual terms into the loss function of the model, including at least data fitting terms, physical residual terms, and process constraint terms;

[0092] The total loss function is:

[0093] ;

[0094] in, Represents the total loss function of the recognition model, Represents the data fitting term, which characterizes the difference between the model prediction results and the actual observed data; represents the physical residual term, corresponding to the residual of the partial differential equation; Represents process constraints, which are constraints imposed by construction technology on progress; , and Represents the adaptive weight parameter, which is used to adjust the relative importance of the data fitting term, physical residual term, and process constraint term in the total loss function;

[0095] The adjustment submodule is used to adjust the adaptive weights to obtain the PINN-Transformer construction progress deviation identification model.

[0096] 104. The construction progress identification module is used to input characteristic water conservancy construction data into the PINN-Transformer construction progress deviation identification model for identification, and use the MAML algorithm to extract common feature representations during the training process to obtain construction progress results.

[0097] Specifically, this embodiment further includes a data input unit for inputting characteristic water conservancy construction data into the PINN-Transformer construction progress deviation identification model for identification and loading model parameters;

[0098] The path processing unit is used to perform dual-path processing at the diagnostic layer in the identification model. The physical path is based on the construction dynamics equation for mechanism deduction, and the data path captures long-term dependencies through the Transformer.

[0099] The progress identification unit is used to generate multi-dimensional identification results, including at least the current progress deviation index, key bottleneck process location and potential risk event warning;

[0100] The meta-learner unit is used to extract common feature representations during training using the MAML algorithm, load the basic feature encoder from the cross-project meta-knowledge base, select the first three days of project data as the support set, freeze the physical constraint layer parameters, adjust the feature extraction layer, and use second-order gradient optimization to accelerate convergence;

[0101] The teacher-student architecture unit is used to construct the teacher-student model architecture, where the teacher model is a complete PINN-Transformer architecture and the student model is a lightweight diagnostic network.

[0102] 1. Multi-source data integration input: align pre-processed feature data according to the spatiotemporal dimensions, including BIM geometric semantics, sensor time series flow, geological parameter matrix, etc.; automatically detect data integrity and trigger missing data interpolation mechanism;

[0103] 2. Physically enhanced diagnosis model inference: Load the pre-trained PINN-Transformer model parameters; implement dual-path processing in the diagnosis layer:

[0104] Physical path: Mechanism deduction based on construction dynamics equations; Data path: Capture long-term dependencies through Transformer. Generate multi-dimensional diagnostic results: Current progress deviation index (0-100%); Key bottleneck process location; Potential risk event warning;

[0105] 3. Meta-learning for rapid adaptation:

[0106] Meta-initialization loading: load the basic feature encoder from the cross-project meta-knowledge base; small sample fine-tuning: select the data of the first three days of the current project as the support set, freeze the parameters of the physical constraint layer, and only adjust the feature extraction layer. Use second-order gradient optimization to accelerate convergence. Elastic weight consolidation: maintain the stability of important parameters during fine-tuning to prevent catastrophic forgetting;

[0107] 4. Dynamic Knowledge Distillation

[0108] Construct the teacher-student model architecture: Teacher model: Complete PINN-Transformer architecture Student model: Lightweight diagnostic network (deployed to edge devices);

[0109] Implement progressive distillation: Phase 1: full feature imitation learning; Phase 2: key decision-making layer attention transfer;

[0110] 5. Diagnostic result generation and verification:

[0111] Output level fusion: physical simulation results (30% weight); data-driven prediction (50% weight); expert experience rules (20% weight); implementation of a triple verification mechanism: digital twin visualization comparison; on-site drone inspection review; and manual confirmation by the supervisor.

[0112] Its beneficial effect is that by integrating BIM models, geological radar and sensor data, it realizes the comprehensive collection of multi-dimensional information of water conservancy construction sites. The segmentation of multimodal data, the mapping of time series data to process state space, and the joint spatiotemporal-physical embedding representation effectively solve the semantic alignment and feature fusion problems of cross-modal data. The PINN-Transformer construction progress deviation identification model has both physical information constraints and the ability to capture long-term time series dependencies. PINN introduces the residual term of the partial differential equation to integrate the physical laws of water conservancy construction into model training to ensure that the recognition results are consistent with the actual engineering physical characteristics; the Transformer module effectively captures the dependencies in long-distance time series data and enhances the model's ability to analyze the complex dynamic changes in the construction progress. Combining the MAML algorithm to extract common feature representations not only improves the efficiency of model training, but also enhances its generalization ability, and can accurately identify construction progress deviations, providing a scientific and reliable decision-making basis for water conservancy project construction management.

[0113] In this embodiment, please refer to Figure 2 In a second embodiment of an intelligent management system for water conservancy project construction progress according to an embodiment of the present invention, the multimodal data acquisition module includes the following submodules:

[0114] The BIM model information extraction submodule is used to use the Revit API secondary development tool to export the geometric parameters of components, including at least three-dimensional coordinates, volume and surface area, and topological relationships; establish a water conservancy project ontology semantic library, including at least functional categories, material categories, and process categories, and use the NLP parser to extract the process parameters in the semantic library to obtain geometric and semantic information;

[0115] The geological radar data acquisition submodule is used to collect real-time scanning information of the geological radar equipment and convert the electromagnetic wave reflection delay and amplitude data in the real-time scanning information into a three-dimensional geological parameter distribution map using the conjugate gradient inversion algorithm;

[0116] The mechanical posture data acquisition submodule is used to obtain real-time data from excavators, tower cranes, and pump trucks. It includes at least a 9-axis IMU, GNSS-RTK, laser rangefinder, stress sensor, flow meter, pressure sensor, mechanical vibration frequency, and current load to obtain mechanical posture data.

[0117] The data preprocessing submodule is used to use the PTP protocol to time-align geometric and semantic information, three-dimensional geological parameter distribution maps, and mechanical posture data, delete outliers from the aligned data, and obtain multimodal water conservancy construction data.

[0118] The beneficial effect lies in the comprehensive collection of multi-dimensional information about water conservancy construction sites by integrating BIM models, geological radar, and sensor data. The geometric and semantic information of the BIM model accurately depicts the structural characteristics of the project, the three-dimensional geological parameter distribution map of the geological radar clearly displays the underground geological conditions, and the mechanical posture data acquired by the sensors reflects the dynamics of the construction machinery in real time. After preliminary processing, the resulting multimodal water conservancy construction data lays the foundation for subsequent analysis.

[0119] In this embodiment, please refer to Figure 3 In a third embodiment of an intelligent management system for water conservancy project construction progress according to an embodiment of the present invention, the data feature extraction module includes the following subunits:

[0120] The semantic segmentation submodule is used to segment the geometric and semantic information in the multimodal water conservancy construction data using the graph attention segmentation network. The graph structure is defined, where the vertex set is the BIM component and the edge set is the physical connection or process dependency of the component, and the segmented data is obtained.

[0121] The state mapping submodule is used to map the mechanical vibration frequency and current load in the multimodal water conservancy construction data into the process state space to obtain cross-modal semantically aligned data;

[0122] The sample generation submodule is used to generate digital twins of the stress and seepage fields in the construction area based on finite element analysis. It encodes the physical field simulation results in the cross-modal semantic alignment data into high-dimensional feature vectors. It also establishes positive and negative sample pairs, where the positive sample is the sensor data and physical field feature vector at the same time and space point, and the negative sample is a random combination of different construction stages or spatial locations.

[0123] The embedding representation submodule is used to train the contrastive learning model to maximize the similarity of positive samples and minimize the similarity of negative samples, and to perform a spatiotemporal-physical joint embedding representation on the cross-modal semantic alignment data to obtain characteristic water conservancy construction data.

[0124] The beneficial effect is that the Transformer module effectively captures the dependencies in long-distance time series data, improving the model's ability to analyze complex dynamic changes in construction progress. Combined with the MAML algorithm to extract common feature representations, it not only improves the model training efficiency but also enhances its generalization ability.

[0125] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention, and such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent management system for water conservancy project construction progress, characterized in that: The water conservancy project construction progress intelligent management system includes the following modules: The multimodal data acquisition module is used to obtain geometric and semantic information in the BIM model, obtain three-dimensional geological parameter distribution maps through geological radar equipment, and use sensors to obtain mechanical posture data of water conservancy construction sites. The acquired data is preliminarily processed to obtain multimodal water conservancy construction data; A data feature extraction module is used to segment the geometric and semantic information in the multimodal water conservancy construction data, map the time series data to the process state space, obtain cross-modal semantic alignment data, and perform a spatiotemporal-physical joint embedding representation on the cross-modal semantic alignment data to obtain characteristic water conservancy construction data; A recognition model building module is used to build a recognition model based on the PINN physical information neural network. A Transformer module is added to the recognition model to capture long-distance temporal dependencies, and a partial differential equation residual term is introduced into the model's loss function to obtain a PINN-Transformer construction progress deviation recognition model. The construction progress identification module is used to input the characteristic water conservancy construction data into the PINN-Transformer construction progress deviation identification model for identification, and use the MAML algorithm to extract common feature representations in the training process to obtain construction progress results.

2. The intelligent management system for water conservancy project construction progress according to claim 1, characterized in that: The multimodal data acquisition module includes the following submodules: The BIM model information extraction submodule is used to export the geometric parameters of components using the Revit API secondary development tool to obtain geometric and semantic information; The geological radar data acquisition submodule is used to collect real-time scanning information of the geological radar equipment and convert the electromagnetic wave reflection delay and amplitude data in the real-time scanning information into a three-dimensional geological parameter distribution map using a conjugate gradient inversion algorithm; The mechanical posture data acquisition submodule is used to obtain real-time data from excavators, tower cranes, and pump trucks to obtain mechanical posture data; The data preprocessing submodule is used to use the PTP protocol to time-align the geometric and semantic information, the three-dimensional geological parameter distribution map and the mechanical posture data, delete outliers from the aligned data, and obtain multimodal water conservancy construction data.

3. The intelligent management system for water conservancy project construction progress according to claim 1, characterized in that: The data feature extraction module includes the following submodules: A semantic segmentation submodule, configured to segment the geometric and semantic information in the multimodal water conservancy construction data using a graph attention segmentation network to obtain segmented data; A state mapping submodule, configured to map the mechanical vibration frequency and current load in the multimodal water conservancy construction data to a process state space to obtain cross-modal semantically aligned data; a sample generation submodule, for generating digital twins of the stress field and seepage field of the construction area based on finite element analysis, and encoding the physical field simulation results in the cross-modal semantic alignment data into high-dimensional feature vectors; The embedding representation submodule is used to train the contrastive learning model to maximize the similarity of positive samples and minimize the similarity of negative samples, and to perform a spatiotemporal-physical joint embedding representation on the cross-modal semantic alignment data to obtain characteristic water conservancy construction data.

4. The intelligent management system for water conservancy project construction progress according to claim 1, characterized in that: The recognition model building module includes the following submodules: Establish a submodule to build an identification model based on the PINN physical information neural network and define the partial differential equation of construction progress evolution; Submodules are added to build a dual-channel input architecture for the recognition model, including a physical channel and a process channel, and a Transformer module is added to the recognition model to capture long-range temporal dependencies.

5. The intelligent management system for water conservancy project construction progress according to claim 1, characterized in that: The recognition model building module also includes the following submodules: Introducing a submodule for introducing partial differential equation residual terms into the loss function of the model, including at least data fitting terms, physical residual terms, and process constraint terms; The adjustment submodule is used to adjust the adaptive weights to obtain the PINN-Transformer construction progress deviation identification model.

6. The intelligent management system for water conservancy project construction progress according to claim 1, characterized in that: The construction progress identification module includes the following units: The data input unit is used to input the characteristic water conservancy construction data into the PINN-Transformer construction progress deviation identification model for identification and load model parameters. A path processing unit is used to perform dual-path processing in the diagnosis layer of the identification model, wherein the physical path is based on the construction dynamics equation for mechanism deduction, and the data path captures long-term dependencies through Transformer; The progress identification unit is used to generate multi-dimensional identification results, including at least the current progress deviation index, key bottleneck process location and potential risk event warning.

7. The intelligent management system for water conservancy project construction progress according to claim 1, characterized in that: The construction progress identification module also includes the following subunits: The meta-learner unit is used to extract common feature representations during training using the MAML algorithm, load the basic feature encoder from the cross-project meta-knowledge base, select the first three days of project data as the support set, freeze the physical constraint layer parameters, adjust the feature extraction layer, and use second-order gradient optimization to accelerate convergence; The teacher-student architecture unit is used to construct the teacher-student model architecture, where the teacher model is a complete PINN-Transformer architecture and the student model is a lightweight diagnostic network.

8. A method for implementing an intelligent management system for water conservancy project construction progress as claimed in claim 1, characterized in that: The method comprises the following steps: Acquire geometric and semantic information from the BIM model, obtain three-dimensional geological parameter distribution maps through geological radar equipment, use sensors to obtain mechanical posture data of the water conservancy construction site, and perform preliminary processing on the acquired data to obtain multimodal water conservancy construction data; Segmenting the geometric and semantic information in the multimodal water conservancy construction data, mapping the time series data to the process state space to obtain cross-modal semantic alignment data, and performing a spatiotemporal-physical joint embedding representation on the cross-modal semantic alignment data to obtain characteristic water conservancy construction data; A recognition model is established based on the PINN physical information neural network. A Transformer module is added to the recognition model to capture long-range temporal dependencies. A partial differential equation residual term is introduced into the model's loss function to obtain the PINN-Transformer construction progress deviation recognition model. The characteristic water conservancy construction data is input into the PINN-Transformer construction progress deviation identification model for identification, and the MAML algorithm is used to extract common feature representations during the training process to obtain the construction progress results.

9. A method for implementing an intelligent management system for water conservancy project construction progress as claimed in claim 8, characterized in that: The method further comprises the following steps: Based on the PINN physical information neural network, an identification model is established to define the partial differential equation of construction progress evolution; A dual-channel input architecture of the recognition model is constructed, including a physical channel and a process channel, and a Transformer module is added to the recognition model to capture long-distance temporal dependencies.

10. A method for implementing an intelligent management system for water conservancy project construction progress as claimed in claim 8, characterized in that: The method further comprises the following steps: Introducing partial differential equation residual terms into the loss function of the model, including at least data fitting terms, physical residual terms, and process constraint terms; The adaptive weights are adjusted to obtain the PINN-Transformer construction progress deviation identification model.

Images