Digital twinning-based refined decoration construction full-process intelligent management and control system

Abstract

The invention discloses a refined decoration construction full-process intelligent management and control system based on digital twinning, and the system comprises a data collection and virtual-real mapping module which is used for generating a virtual-real mapping database; the improved PIML prediction module is used for outputting a prediction result; the physical constraint attention fusion layer module is used for generating fusion feature representation and a physical constraint residual signal; the physical residual gating unit module is used for marking an update path; the GANopen reconstruction and anomaly detection module is used for generating a reconstruction feature vector and reconstruction residual error distribution; the abnormal score and residual signal feedback module is used for forming a residual signal; the parameter updating interface module is used for covering virtual model node parameters; and the three-dimensional visualization display module is used for generating a three-dimensional visualization result. The method is suitable for a complex building construction full-process intelligent management and control scene driven by multi-source heterogeneous data.

Description

Technical Field

[0001] This invention relates to the field of building information technology and intelligent construction technology, and in particular to an intelligent management and control system for the entire process of interior decoration construction based on digital twins. Background Technology

[0002] In the current field of building decoration engineering, the construction process of fine decoration often faces multiple challenges due to the large number of overlapping procedures, multi-disciplinary collaboration, and high reliance on manual experience. These challenges include uncontrollable construction progress, delayed discovery of quality problems, and waste of material resources. To address these issues, BIM (Building Information Modeling) technology is gradually being widely used in construction management. Using BIM models for visual management and process coordination of the construction process has become an effective means. At the same time, with the deployment of sensor networks, video surveillance, and smart terminals, the data collection capabilities of construction sites have been significantly improved, forming a foundation for construction process perception based on multi-source heterogeneous data.

[0003] In recent years, digital twin technology has been gradually applied to the field of engineering construction. By constructing a virtual model synchronized with the physical construction site, it achieves real-time mapping of construction status and process simulation, becoming one of the important paths to realize intelligent construction. Some studies have attempted to combine digital twins with BIM and sensor data to monitor and provide early warnings for construction sites; some solutions have also introduced machine learning models to predict and analyze construction progress or quality. However, there are still significant shortcomings in existing technologies: on the one hand, physical laws have not been effectively integrated into the modeling process, resulting in insufficient model response to physical constraints, leading to problems such as prediction distortion or constraint violation; on the other hand, anomaly detection mostly relies on static rules or error thresholds, which cannot be flexibly adjusted and dynamically updated according to actual construction conditions. In addition, there is usually a lack of effective parameter interaction and feedback mechanisms between the digital twin system and the prediction model, making it difficult to correct prediction deviations in a timely manner and reflect them in the 3D visualization results, thus affecting the decision-making efficiency and on-site response capabilities of construction management personnel.

[0004] Therefore, how to provide an intelligent management and control system for the entire process of interior decoration construction based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent management and control system for the entire process of interior decoration construction based on digital twins. This invention fully integrates BIM modeling, physical constraint perception, improved machine learning prediction and anomaly detection algorithms, and describes in detail the closed-loop process from data acquisition, virtual-real mapping, intelligent prediction, anomaly detection, parameter updating and 3D visualization. It realizes intelligent monitoring, risk identification and dynamic control of the entire interior decoration construction process, and has the advantages of high prediction accuracy, fast response speed and high degree of intelligent construction management.

[0006] According to an embodiment of the present invention, an intelligent management and control system for the entire process of interior decoration construction based on digital twins includes: The data acquisition and virtual-real mapping module is used to establish construction nodes and virtual model nodes based on BIM, collect multi-source data and generate a virtual-real mapping database. The improved PIML prediction module is used to receive data-driven features and physical constraint features, output prediction results and write them to the parameter mapping table; The physically constrained attention fusion layer module is used to generate fused feature representations and physically constrained residual signals during forward propagation. The physical residual gating unit module is used to mark the update path and write it into the parameter register area based on the gradient direction and residual magnitude during backpropagation. The GANomaly reconstruction and anomaly detection module is used to reconstruct the prediction results and measured data, and generate the reconstruction feature vector and reconstruction residual distribution. An anomaly scoring and residual signal feedback module is used to generate an anomaly scoring vector from the reconstructed residual distribution, compare thresholds, and form a residual signal. The parameter update interface module is used to send and receive parameters and overwrite virtual model node parameters; The 3D visualization module is used to generate 3D visualization results.

[0007] Optionally, modules can be integrated using the following methods: Step 1: Construct a digital twin model of the interior decoration construction scene, collect multi-source data, preprocess the multi-source data, and establish a virtual-real mapping database; Step 2: Input the feature set extracted from the virtual-real mapping database into the improved PIML prediction model, set up a physical constraint attention fusion layer and a physical residual gating unit. In the forward propagation stage, the physical constraint attention fusion layer generates fused feature representations based on each physical constraint residual. In the back propagation stage, the physical residual gating unit adjusts the update path of the data branch and the physical branch according to the gradient direction and the residual magnitude. Step 3: Perform joint training and inference on the feature set, output the prediction results, and write the prediction results into the parameter mapping table of the digital twin model; Step 4: Input the prediction results and the actual measured data at the construction site into the GANomaly model. By reconstructing the prediction results and the measured data, the reconstructed feature vector and the reconstructed residual distribution are obtained. Step 5: Calculate the anomaly score corresponding to the reconstructed residual distribution, determine the degree of deviation based on the preset threshold, and generate a residual signal when the anomaly score exceeds the threshold. Feedback is sent to the parameter update interface of the improved PIML prediction model and the digital twin model to synchronously correct the physical constraint parameters and virtual model node parameters. Step 6: Update the initial state set and timing information in the corrected digital twin model to generate a 3D visualization result.

[0008] Optionally, the feature is that the construction of the digital twin model of the interior decoration construction scene involves collecting multi-source data, preprocessing the multi-source data, and establishing a virtual-real mapping database, specifically as follows: Construct a digital twin model of the interior decoration construction scenario, and define construction units and divide construction nodes based on the BIM model; Assign a unique identifier and spatial coordinates to each construction node, establish construction dependencies between nodes in the digital twin model, and generate a corresponding virtual model node for each construction node in the digital twin model; Collect multi-source data from BIM models, environmental sensors, construction logs, and video surveillance, and perform timestamp alignment and feature standardization processing on the multi-source data; Based on data attributes, the preprocessed multi-source data is divided into data-driven features and physical constraint features; An initial state set containing data-driven features and physical constraint features is generated for each construction node, and a data mapping relationship is established using the construction node index as the association key. Synchronously record the timing information of construction nodes, including the start time of construction, the planned completion time, and the order of dependencies; Construct a virtual-physical mapping database that includes an initial state set, data-driven features, physical constraint features, construction node indexes, and time sequence information.

[0009] Optionally, the physical constraint attention fusion layer generates a fusion feature representation based on the residuals of each physical constraint during the forward propagation stage, specifically as follows: Read the feature set corresponding to each construction node from the virtual-real mapping database, and divide the channel according to data-driven features and physical constraint features; Input the data-driven features into the data-driven branch of the improved PIML prediction model, and input the physical constraint features into the physical constraint branch of the improved PIML prediction model to generate the corresponding branch feature representations. During the forward propagation phase, the physically constrained attention fusion layer is invoked; Perform linear mapping on data-driven branch features to generate query vectors, and perform linear mapping on physical constraint branch features to generate key vectors and value vectors; The attention weight matrix is ​​calculated based on the similarity between the query vector and the key vector, and then normalized. The normalized attention weight matrix and the value vector are weighted and calculated to generate a fused feature representation; The fusion feature representation is matched with the index table corresponding to the construction node identifier, and written into the fusion result buffer in the order of construction nodes; Complete the forward propagation calculation for the current batch, calculate the difference between the physical constraint prediction value output by the physical constraint branch and the result of the preset physical equation, and generate the physical constraint residual signal as an intermediate signal.

[0010] Optionally, the physical residual gating unit adjusts the update paths of the data branch and the physical branch according to the gradient direction and residual magnitude during the backpropagation phase, specifically as follows: During the backpropagation phase, the physical residual gating unit is invoked; Perform angle calculation on the gradient directions of the data-driven branch and the physical constraint branch to generate directional relationship identifiers; Read the physical constraint residual signal, perform a threshold comparison operation on the residual amplitude, and generate a residual amplitude identifier; The residual amplitude is the magnitude of the physically constrained residual signal; The gating control unit generates a gating status signal based on the direction relationship identifier and the residual magnitude identifier, and marks the update path of the data-driven branch and the physical constraint branch; After all construction node indexes have been processed, the updated path results are written to the parameter register of the improved PIML prediction model to complete the backpropagation calculation for the current batch.

[0011] Optionally, the step of performing joint training and inference on the feature set, outputting prediction results, and writing the prediction results into the parameter mapping table of the digital twin model specifically involves: Set the joint training round index and batch size; In the training path, the fused feature representation is fed into the backbone of the improved PIML prediction model to generate data-driven branch output and physical constraint branch output respectively. Construct a training objective consisting of a data prediction error term and a physical consistency constraint term, perform parameter updates, and record the training round index; During the parameter update process, the parameter update interface of the improved PIML prediction model is called, and the updated model parameters are written into the parameter update interface of the digital twin model. Under the inference path, the fusion feature representation is forward-computed based on the latest parameters, and the prediction results are output, including the predicted value of construction progress, quality risk index and physical constraint. The prediction results are written into a parameter mapping table according to the construction node index and time index; Synchronize new records in the parameter mapping table to the virtual-real mapping database, and update the current snapshot of the initial state set and timing information corresponding to the construction node.

[0012] Optionally, the step of inputting the prediction results and the measured data from the construction site into the GANomaly model, and generating the reconstructed feature vector and the reconstructed residual distribution by reconstructing the prediction results and the measured data, specifically involves: Read the actual construction site data corresponding to the construction node index and time index from the virtual-real mapping database. The actual construction site data includes progress completion rate, quality inspection parameters and physical response parameters. The prediction results are aligned with the actual measured data at the construction site, and the channels are spliced ​​to generate the input tensor of the GANomaly model and establish a one-to-one correspondence with the construction node index. The encoder module of the GANomaly model is invoked to generate a latent representation vector from the input tensor and write it to the reconstruction buffer. The decoder module of the GANomaly model is invoked to perform reconstruction operations on the latent representation vector and output the reconstructed output vector. Perform element-wise difference operations between the input tensor and the reconstructed output vector to generate the reconstructed feature vector and the reconstructed residual vector; The reconstruction residual vectors are grouped and statistically analyzed according to the construction node index and time index to generate the reconstruction residual distribution; The reconstructed feature vector and the reconstructed residual distribution are written into the anomaly detection result buffer, and an index mapping is established based on the construction node index.

[0013] Optionally, the method is characterized in that the anomaly score corresponding to the calculated reconstructed residual distribution is used to determine the degree of deviation based on a preset threshold. When the anomaly score exceeds the threshold, a residual signal is generated and fed back to the parameter update interface of the improved PIML prediction model and the digital twin model to synchronously correct the physical constraint parameters and virtual model node parameters. Specifically: Anomaly score vectors are generated by reconstructing the residual distribution; Perform a threshold comparison operation on the abnormal score vectors based on the preset score threshold to generate a deviation status indicator; When the deviation status is marked as exceeding the limit, a residual signal containing the construction node index, time index, anomaly score vector and reconstruction residual distribution is constructed. The residual signal is written into the update queue of the parameter update interface of the improved PIML prediction model and the parameter update interface of the digital twin model. The parameter group is located based on the residual signal, and parameter overwrite is performed in the parameter register area of ​​the improved PIML prediction model. The construction nodes are located based on the residual signals, and the node parameters of the virtual model are overwritten. Write the abnormal scoring vector, deviation status identifier, residual signal and parameter update results of this batch into the correction log area, and synchronize the updated virtual model node parameters to the virtual-real mapping database.

[0014] Optionally, the step of updating the initial state set and temporal information in the corrected digital twin model to generate a three-dimensional visualization result specifically involves: The updated virtual model node parameters, initial state set, and timing information are read from the parameter update interface of the digital twin model and the virtual-real mapping database, and a batch of state snapshots are generated according to the construction node index and time index. Write the virtual model node parameters into the node data area of ​​the digital twin model to update the timing information of the construction node; Write the updated initial state set and time series information into the virtual-real mapping database, and append the current snapshot record according to the time index; Load the geometric data and virtual model node set of the BIM model, and establish an index mapping table from the construction node index to the three-dimensional scene entity; Based on the parameter mapping table and virtual model node parameters, the rendering attribute set and annotation information are assigned to the 3D scene entities, and the rendering attribute set is bound to the construction node index using key-value pairs. Perform frame-level rendering on 3D scene entities according to time index, and generate a sequence of 3D visualization results corresponding to the time index; Write the 3D visualization result sequence, index mapping table, and current batch status snapshot to the visualization result storage area, and write the recorded timestamps to the display log area to complete the 3D visualization result output.

[0015] The beneficial effects of this invention are: This invention constructs an intelligent management and control system for the entire process of interior decoration construction based on digital twins. It achieves data-driven and physical constraint-integrated modeling of the entire construction process and forms a closed-loop control mechanism from multi-source data acquisition and predictive analysis to anomaly feedback and parameter updates. First, by dividing construction nodes into virtual model nodes using a BIM model, this invention establishes a mapping relationship between the construction site and the digital space, effectively improving the organization efficiency and semantic consistency of construction data. Second, by leveraging an improved PIML prediction model, which integrates data-driven and physical constraint branches, and utilizes an attention fusion layer and gating units to achieve more physically accurate intelligent predictions, significantly improving the accuracy of construction progress and quality risk assessment. Furthermore, the introduction of a GANomaly model enables reconstruction and residual analysis of the predicted results and measured data, further enhancing the accuracy and responsiveness of anomaly detection. It can also dynamically adjust model parameters and virtual node states based on the reconstructed residuals, completing real-time correction of the construction process. Finally, through the 3D visualization module, the predictive analysis results and node status are presented in a visual form, providing construction managers with an intuitive, dynamic, and interactive decision support tool, and comprehensively improving the intelligent management level and construction quality control capabilities of the interior decoration project. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of an intelligent management and control system for the entire process of interior decoration construction based on digital twins, as proposed in this invention. Figure 2 This is an overall flowchart of an intelligent management and control method for the entire process of interior decoration construction based on digital twins, as proposed in this invention. Figure 3 This is a schematic diagram of the improved PIML prediction model structure of an intelligent management and control system for the entire process of interior decoration construction based on digital twins, as proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figure 1-3 A digital twin-based intelligent management and control system for the entire process of interior decoration construction, comprising: The data acquisition and virtual-real mapping module is used to establish construction nodes and virtual model nodes based on BIM, collect multi-source data and generate a virtual-real mapping database. The improved PIML prediction module is used to receive data-driven features and physical constraint features, output prediction results and write them to the parameter mapping table; The physically constrained attention fusion layer module is used to generate fused feature representations and physically constrained residual signals during forward propagation. The physical residual gating unit module is used to mark the update path and write it into the parameter register area based on the gradient direction and residual magnitude during backpropagation. The GANomaly reconstruction and anomaly detection module is used to reconstruct the prediction results and measured data, and generate the reconstruction feature vector and reconstruction residual distribution. An anomaly scoring and residual signal feedback module is used to generate an anomaly scoring vector from the reconstructed residual distribution, compare thresholds, and form a residual signal. The parameter update interface module is used to send and receive parameters and overwrite virtual model node parameters; The 3D visualization module is used to generate 3D visualization results.

[0019] This invention constructs a system architecture consisting of a data acquisition and virtual-real mapping module, an improved PIML prediction module, a physical constraint attention fusion layer module, a physical residual gating unit module, a GANomaly reconstruction and anomaly detection module, an anomaly scoring and residual signal feedback module, a parameter update interface module, and a 3D visualization display module. This forms an intelligent closed-loop management and control system that integrates data acquisition, model prediction, anomaly detection, parameter correction, and visualization display. It realizes virtual-real fusion modeling, physical constraint adaptive learning, and dynamic visualization management of construction status during the fine decoration construction process.

[0020] In this embodiment, the modules are interconnected using the following method: Step 1: Construct a digital twin model of the interior decoration construction scene, collect multi-source data, preprocess the multi-source data, and establish a virtual-real mapping database; Step 2: Input the feature set extracted from the virtual-real mapping database into the improved PIML prediction model, set up a physical constraint attention fusion layer and a physical residual gating unit. In the forward propagation stage, the physical constraint attention fusion layer generates fused feature representations based on each physical constraint residual. In the back propagation stage, the physical residual gating unit adjusts the update path of the data branch and the physical branch according to the gradient direction and the residual magnitude. Step 3: Perform joint training and inference on the feature set, output the prediction results, and write the prediction results into the parameter mapping table of the digital twin model; Step 4: Input the prediction results and the actual measured data at the construction site into the GANomaly model. By reconstructing the prediction results and the measured data, the reconstructed feature vector and the reconstructed residual distribution are obtained. Step 5: Calculate the anomaly score corresponding to the reconstructed residual distribution, determine the degree of deviation based on the preset threshold, and generate a residual signal when the anomaly score exceeds the threshold. Feedback is sent to the parameter update interface of the improved PIML prediction model and the digital twin model to synchronously correct the physical constraint parameters and virtual model node parameters. Step 6: Update the initial state set and timing information in the corrected digital twin model to generate a 3D visualization result.

[0021] This invention constructs a digital twin model of the interior decoration construction scenario, collects and preprocesses multi-source data to establish a virtual-real mapping database, and combines an improved PIML prediction model with a physical constraint attention fusion layer and a physical residual gating unit to achieve fusion feature modeling and adaptive parameter updates. After the prediction is completed, the predicted and measured values ​​are reconstructed using a GANomaly model to generate a residual distribution and calculate anomaly scores. If the score exceeds the limit, a residual signal is fed back to trigger the synchronous correction of model parameters and virtual model nodes. Finally, the state information is updated in the corrected digital twin model and a three-dimensional visualization result is generated, forming a closed-loop process for intelligent management and control of interior decoration construction that is data-driven and physically constrained.

[0022] In this embodiment, the construction of a digital twin model of the interior decoration construction scene involves collecting multi-source data, preprocessing the multi-source data, and establishing a virtual-real mapping database, specifically as follows: Construct a digital twin model of the interior decoration construction scenario, and define construction units and divide construction nodes based on the BIM model; Each construction node is assigned a unique identifier and spatial coordinates. Construction dependencies between nodes are established in the digital twin model, and a corresponding virtual model node is generated for each construction node in the digital twin model to represent the virtual entity state of the construction unit. Collect multi-source data from BIM models, environmental sensors, construction logs, and video surveillance, and perform timestamp alignment and feature standardization processing on the multi-source data; Based on data attributes, the preprocessed multi-source data is divided into data-driven features and physical constraint features; the data-driven features are used to characterize the progress status parameters and quality status parameters of the construction nodes, and the physical constraint features are used to characterize the environmental status parameters and physical response parameters of the construction nodes. An initial state set containing the data-driven features and physical constraint features is generated for each construction node, and a node-level data mapping relationship is established using the construction node index as the association key. Synchronously record the timing information of construction nodes, including the start time of construction, the planned completion time, and the order of dependencies; Construct a virtual-physical mapping database that includes an initial state set, data-driven features, physical constraint features, construction node indexes, and time sequence information.

[0023] This invention constructs a digital twin model of a high-end interior decoration construction scenario. Based on the BIM model, it defines construction units and divides them into construction nodes. A corresponding virtual model node is generated for each construction node in the model to represent the virtual entity state of the construction unit. Data from multiple sources, including the BIM model, environmental sensors, construction logs, and video surveillance, is collected. Timestamp alignment and feature standardization are performed, and the data is categorized by attributes into data-driven features representing construction progress and quality status, and physical constraint features representing environmental and physical response status. An initial state set and node-level data mapping relationship are established based on the construction node index, and time-series information is recorded synchronously. Finally, a virtual-real mapping database containing the state characteristics and time-series structure of the entire construction process is generated, providing unified data support and semantic association for subsequent intelligent prediction and dynamic correction.

[0024] In this embodiment, the physical constraint attention fusion layer generates a fusion feature representation based on the residuals of each physical constraint during the forward propagation stage, specifically as follows: Read the feature set that corresponds one-to-one with the construction node from the virtual-real mapping database, divide the channel according to the data-driven features and physical constraint features, and keep the construction node identifier and time index consistent; Input the data-driven features into the data-driven branch of the improved PIML prediction model, and input the physical constraint features into the physical constraint branch of the improved PIML prediction model to generate the corresponding branch feature representations. During the forward propagation phase, the physically constrained attention fusion layer is invoked; Perform dimension alignment and channel mapping on data-driven branch features and physical constraint branch features; perform linear mapping on data-driven branch features to generate query vectors; and perform linear mapping on physical constraint branch features to generate key vectors and value vectors. The attention weight matrix is ​​calculated based on the similarity between the query vector and the key vector, and then normalized. The normalized attention weight matrix and the value vector are weighted and calculated to generate a fused feature representation; The fusion feature representation is matched with the index table corresponding to the construction node identifier, and written into the fusion result buffer according to the order of the construction nodes to establish a one-to-one correspondence between the fusion feature representation and the construction node identifier. The fused features are output to the backbone of the improved PIML prediction model to complete the forward propagation calculation for the current batch; The difference between the physical constraint prediction value output by the physical constraint branch and the result of the preset physical equation is calculated, and the physical constraint residual signal is generated as an intermediate signal.

[0025] This invention introduces a physical constraint attention fusion layer into an improved PIML prediction model, achieving deep fusion of data-driven features and physical constraint features during the forward propagation stage. The system first extracts feature sets corresponding to construction nodes from a virtual-real mapping database, dividing the input data-driven branch and the physical constraint branch by channel, generating branch feature representations for each. Subsequently, the fusion layer performs dimensional alignment and channel mapping on the two types of features, calculates the similarity between the query vector and the key vector based on an attention mechanism, obtains a normalized attention weight matrix, and weights it with the value vector to generate a fused feature representation. After establishing a correspondence between the fusion result and the construction node index table, it is output to the main path for subsequent calculations. Simultaneously, it generates a physical constraint residual signal from the difference between the predicted physical constraint value and the physical equation result, providing a basis for backpropagation and parameter correction, thus achieving effective collaborative modeling of data features and physical constraint information.

[0026] In this embodiment, the physical residual gating unit adjusts the update paths of the data branch and the physical branch according to the gradient direction and residual magnitude during the backpropagation phase, specifically as follows: During the backpropagation phase, the physical residual gating unit is invoked; Write the data-driven branch gradient and the physical constraint branch gradient into the gated buffer and keep the construction node index consistent; Perform angle calculation on the vector directions of the data-driven branch gradient and the physical constraint branch gradient to generate directional relationship identifiers; Read the physical constraint residual signal, perform a threshold comparison operation on the residual amplitude, and generate a residual amplitude identifier; The residual amplitude is the magnitude of the physically constrained residual signal; The gating control unit generates a gating status signal based on the direction relationship identifier and the residual magnitude identifier, and marks the update path of the data-driven branch and the physical constraint branch; After all construction node indexes have been processed, the updated path results are written to the parameter register of the improved PIML prediction model to complete the backpropagation calculation for the current batch.

[0027] This invention introduces a physical residual gating unit into an improved PIML prediction model, enabling dynamic adjustment of the update paths for data-driven and physical constraint branches during the backpropagation phase. Specifically, the system first invokes the gating unit to write the gradients of each branch into a buffer, maintaining consistency in the construction node indices. Then, it calculates the angle between gradient vectors to obtain a directional relationship identifier, and combines this with a threshold judgment of the magnitude of the physical constraint residual signal to generate a residual magnitude identifier. The gating control unit combines these two identifiers to generate a gating state signal, used to determine the parameter update priority and path label for each branch at the current node. Finally, the generated update path information is written into the parameter register to guide the selective updating of subsequent model parameters, thereby improving the model's ability to maintain consistency with physical constraints and its dynamic adaptability.

[0028] In this embodiment, marking the update paths of the data-driven branch and the physical constraint branch specifically involves: Path marker bits are assigned to the data-driven branch and the physical constraint branch respectively. The path marker bits include an enable marker bit and a disable marker bit. A gate control status register area is set up in the gate control control unit to receive gate control status signals and generate a branch update mask matrix. The enable flag will be written to the main diagonal cell of the branch update mask matrix of the corresponding branch, and the disable flag will be written to the zero cell of the branch update mask matrix of the corresponding branch. Build an update path table, using the construction node index as the row key and the branch type as the column key; the branch types are data-driven branches and physical constraint branches. Based on the construction node index order, the branch update mask matrix is ​​matched with the update path table by key value to generate branch update path records; Write the branch update path record into the parameter update scheduling queue, and generate gradient update masks for the corresponding data-driven branch and physical constraint branch respectively. The gradient update mask is sent to the parameter update unit. The enable flag is set for the branch corresponding to the enable flag, and the delay flag is set for the branch corresponding to the disable flag. Write the gating status signal, branch update mask matrix, and branch update path record of this batch into the gating log area to complete the marking.

[0029] This invention achieves branch update control during the backpropagation phase by setting independent path marking mechanisms for data-driven branches and physically constrained branches in an improved PIML prediction model. Specifically, the system assigns enable and disable flags to each branch and establishes a gating state register in the gating control unit, generating a branch update mask matrix based on the gating state signals. The update path table is constructed based on the construction node index and branch type, generating update path records using mask matrix matching and writing them into the parameter update scheduling queue, thereby controlling the gradient update permissions of each branch. Enabling the flag drives normal parameter updates, while disabling the flag temporarily suspends gradient propagation for the corresponding branch, effectively avoiding interference from physical inconsistencies in model training. The system archives relevant gating states and path records to the gating log area, providing a traceable mechanism for dynamic update control during model training.

[0030] In this embodiment, the step of performing joint training and inference on the feature set, outputting prediction results, and writing the prediction results into the parameter mapping table of the digital twin model specifically involves: Read the fused feature representation and physical constraint residual signal, and set the training round index and batch size; In the training path, the fused feature representation is fed into the backbone of the improved PIML prediction model to generate data-driven branch output and physical constraint branch output respectively. Construct a training objective consisting of a data prediction error term and a physical consistency constraint term, perform parameter updates, and record the training round index; The parameter updates include the weight parameters and bias parameters of the data-driven branch, the constraint term weight coefficients and residual correction coefficients in the physical constraint branch, the attention weight matrix of the physical constraint attention fusion layer, the gating threshold parameters and path selection coefficients, and the parameters of the fully connected layer and mapping layer in the backbone path.

[0031] During the parameter update process, the parameter update interface of the improved PIML prediction model is called to write the updated model parameters into the parameter update interface of the digital twin model; a synchronous mapping relationship between the model parameters and the virtual construction node parameters is established. The model parameters include construction progress prediction parameters, quality risk parameters, physical constraint-related parameters, and time-series state parameters for each construction node.

[0032] Under the inference path, the fusion feature representation is forward-computed based on the latest parameters, and the prediction results are output, including the predicted value of construction progress, quality risk index and physical constraint. The prediction results are written into a parameter mapping table according to the construction node index and time index to establish a one-to-one correspondence between the prediction results and the construction node identifiers. The newly added records in the parameter mapping table are synchronized to the virtual-real mapping database, and the current snapshot of the initial state set and time series information corresponding to the construction node is updated. The newly added records are the record items generated in the current training batch and appended to the parameter mapping table, including the construction node index, time index, construction progress prediction value, quality risk index and physical constraint prediction value, which are used to represent the latest prediction results of the model under the current batch inference.

[0033] Archive the model parameters, parameter mapping table update records, and timestamps generated during this batch of training and inference to complete the joint training and inference process.

[0034] After training and inference are completed, the latest parameters are sent to the parameter update interface of the digital twin model through the parameter update interface of the improved PIML prediction model, and the parameter synchronization results are recorded and written to the parameter register area to complete the synchronization of virtual and real parameters.

[0035] This invention achieves synchronized parameter updates between an improved PIML prediction model and a digital twin model by performing joint training and inference operations on the fused feature representation and the physical constraint residual signal. In the training path, the fused feature representation is sequentially fed into the data-driven branch and the physical constraint branch to construct a training objective combining data prediction error terms and physical consistency constraints, updating key model parameters including branch weights, the fusion layer attention matrix, gating thresholds, and backbone mapping parameters. The parameter update interface is called to write the updated model parameters into the digital twin model and establish a one-to-one mapping relationship with the virtual construction node parameters. In the inference path, forward calculations are performed based on the latest model parameters, outputting predicted construction progress values, quality risk indicators, and physical constraint prediction values, which are written into the parameter mapping table using the construction node index and time index as keys, updating the node status snapshot in the virtual-real mapping database. The parameter update results and timestamps generated in this batch are archived, completing the entire process of parameter synchronization and virtual-real linkage through the parameter update interface.

[0036] In this embodiment, the step of inputting the prediction results and the measured data from the construction site into the GANomaly model, and then reconstructing the model by analyzing the prediction results and the measured data to obtain the reconstructed feature vector and the reconstructed residual distribution, specifically involves: Read the actual construction site data corresponding to the construction node index and time index from the virtual-real mapping database. The actual construction site data includes progress completion rate, quality inspection parameters and physical response parameters. The prediction results are aligned with the actual measured data at the construction site, and the channels are spliced ​​to generate the input tensor of the GANomaly model and establish a one-to-one correspondence with the construction node index. The encoder module of the GANomaly model is invoked to generate a latent representation vector from the input tensor and write it to the reconstruction buffer. The decoder module of the GANomaly model is invoked to perform reconstruction operations on the latent representation vector and output the reconstructed output vector. Perform element-wise difference operations between the input tensor and the reconstructed output vector to generate the reconstructed feature vector and the reconstructed residual vector; The reconstruction residual vectors are grouped and statistically analyzed according to the construction node index and time index to generate the reconstruction residual distribution; The reconstructed feature vector and the reconstructed residual distribution are written into the anomaly detection result buffer, and an index mapping is established based on the construction node index.

[0037] This invention effectively identifies and represents abnormal states in interior decoration construction scenarios by jointly reconstructing the output of an improved PIML prediction model with measured data from the construction site using a GANomaly model. Specifically, the system first extracts measured data corresponding to construction node indices and time indices from a virtual-real mapping database, including progress completion rate, quality inspection parameters, and physical response parameters. This data is then aligned with the execution time of the prediction results and concatenated with the channels to construct the input tensor of the GANomaly model. Subsequently, the input tensor is encoded into a latent representation by an encoder and then decoded to reconstruct the output vector. The system performs element-wise difference operations on the original input tensor and the reconstructed output vector to generate reconstructed feature vectors and reconstructed residual vectors. These are further statistically analyzed according to construction node and time indices to form a reconstructed residual distribution. Finally, the reconstructed features and residual information are written into the anomaly detection result buffer, establishing a mapping relationship with the construction node indices, providing a basis for subsequent anomaly scoring and parameter updates.

[0038] In this embodiment, the calculation of the anomaly score corresponding to the reconstructed residual distribution is performed. A preset threshold is used to determine the degree of deviation. When the anomaly score exceeds the threshold, a residual signal is generated and fed back to the parameter update interface of the improved PIML prediction model and the digital twin model. This synchronously corrects the physical constraint parameters and virtual model node parameters. Specifically: Anomaly score vectors are generated by reconstructing the residual distribution. The specific process of generating anomaly score vectors is as follows: normalization and segmented statistical operations are performed on the reconstructed residual distribution, and residual samples are aggregated according to construction node index and time index; the anomaly score calculation unit is called to calculate the central tendency index and dispersion index of the reconstructed residual samples of each construction node; the central tendency index and dispersion index are combined into anomaly score values ​​according to weights, anomaly score vectors are generated and written into the anomaly score buffer.

[0039] Perform a threshold comparison operation on the abnormal score vectors based on the preset score threshold to generate a deviation status indicator; When the deviation status is marked as exceeding the limit, a residual signal containing the construction node index, time index, anomaly score vector and reconstruction residual distribution is constructed. The residual signal is written into the update queue of the parameter update interface of the improved PIML prediction model and the parameter update interface of the digital twin model. The parameter group is located based on the residual signal, and parameter overwrite is performed in the parameter register of the improved PIML prediction model; the construction node index and time index are read from the residual signal, and the parameter record corresponding to the index is retrieved in the parameter register; based on the index position of the progress component, quality component and physical component in the anomaly scoring vector, the corresponding parameter block in the data-driven branch and physical constraint branch is matched; the matched parameter block is identified as a parameter group and written to the parameter overwrite queue.

[0040] The construction nodes are located based on the residual signals, and the node parameters of the virtual model are overwritten. Write the abnormal scoring vector, deviation status identifier, residual signal and parameter update results of this batch into the correction log area, and synchronize the updated virtual model node parameters to the virtual-real mapping database.

[0041] This invention introduces an anomaly scoring mechanism based on the reconstruction residual distribution, enabling intelligent identification and adaptive parameter correction of deviations in interior decoration construction scenarios. Specifically, the system first normalizes and performs segmented statistical processing on the reconstruction residual distribution, aggregating residual samples for each construction node under a specific time index, calculating their central tendency and dispersion indices, and fusing them to generate an anomaly scoring vector. When the anomaly score exceeds a preset threshold, the system constructs a residual signal containing node index, time index, residual information, and scoring indices, and feeds it back to the parameter update interface of the improved PIML prediction model and digital twin model. Within the model, by analyzing the progress, quality, and physical constraint components in the anomaly scoring vector, the system locates the parameter groups requiring correction in the data-driven and physical constraint branches, performing an overwrite operation. Simultaneously, the parameters of the corresponding virtual model nodes are also corrected and written to the virtual-real mapping database, thereby achieving consistency maintenance and dynamic updates between the virtual and real models. All correction operations and results are ultimately archived in the correction log area, ensuring the system's prediction accuracy and the dynamic reliability of construction modeling.

[0042] In this embodiment, updating the initial state set and temporal information in the modified digital twin model to generate a three-dimensional visualization result specifically involves: The updated virtual model node parameters, initial state set, and timing information are read from the parameter update interface of the digital twin model and the virtual-real mapping database, and a batch of state snapshots are generated according to the construction node index and time index. Write the virtual model node parameters into the node data area of ​​the digital twin model to update the timing information of the construction node; Write the updated initial state set and time series information into the virtual-real mapping database, and append the current snapshot record according to the time index; Load the geometric data and virtual model node set of the BIM model, and establish an index mapping table from the construction node index to the three-dimensional scene entity; Based on the parameter mapping table and virtual model node parameters, the rendering attribute set and annotation information are assigned to the 3D scene entities, and the rendering attribute set is bound to the construction node index using key-value pairs. Perform frame-level rendering on 3D scene entities according to time index, and generate a sequence of 3D visualization results corresponding to the time index; Write the 3D visualization result sequence, index mapping table, and current batch status snapshot to the visualization result storage area, and write the recorded timestamps to the display log area to complete the 3D visualization result output.

[0043] This invention achieves 3D visualization output of the interior decoration construction process by updating the initial state set and timing information of construction nodes in a corrected digital twin model. Specifically, it includes: reading corrected virtual model node parameters and timing data from a parameter update interface and a virtual-real mapping database to generate a state snapshot of the current batch; updating the node data area and timing information, loading the BIM model's geometric information and establishing a one-to-one mapping relationship between construction nodes and 3D scene entities; assigning rendering attributes and annotation information to each 3D entity based on a parameter mapping table, and performing frame-level rendering according to a time index to form a sequence of 3D visualization results covering the entire construction process; finally, writing the visualization results, state snapshots, and index mapping table into a storage area and recording the display timestamp, constructing a dynamic and traceable construction visualization, thus improving the timing expressiveness and interactive intuitiveness of interior decoration construction modeling. Example

[0044] To verify the feasibility of this invention in practice, it was applied to the intelligent management and control of the entire process of interior decoration construction for a high-rise residential building. This project included multiple construction units, such as wall finishing, floor tile laying, ceiling installation, water and electricity pipeline installation, and equipment commissioning, with a construction period of 120 days and 54 construction milestones. In traditional construction processes, project management relies heavily on manual inspections and experience-based scheduling, which can easily lead to delays, uneven construction quality, and poor energy consumption control. Especially during the cross-disciplinary construction phases, the lack of precise data support between different processes results in delayed construction status assessments, leading to untimely on-site resource allocation, high rework rates, and significant schedule deviations. This invention, by constructing a digital twin model, achieves real-time mapping and dynamic feedback between the virtual model and the actual construction site, effectively solving the problems of invisible status during construction, untimely decision-making, and difficulty in quantifying quality.

[0045] In practical applications, the entire construction area is first modeled in 3D based on the BIM model. Construction units and virtual model nodes are defined according to the design zones, and each node is assigned a unique identifier and spatial coordinates. The system collects data from multiple sensors deployed on-site, including environmental parameters such as temperature and humidity, dust concentration, noise level, and vibration amplitude, as well as information from construction logs and video monitoring systems, such as work progress, worker attendance, and equipment operating status. All collected data is aggregated into the virtual-physical mapping database at 5-minute intervals and then standardized and timestamped to generate a structured feature set. During the construction period from May to July 2025, the system collected approximately 16 million data entries from multiple sources, with an average of approximately 270,000 new data entries added daily.

[0046] During the model training phase, the aforementioned feature set is input into the improved PIML prediction model. Data-driven features describe the progress and quality status, while physical constraint features characterize environmental conditions and construction response. Through a physical constraint attention fusion layer, the model can adaptively adjust the weights of each feature dimension during forward propagation, generating a fused feature representation and achieving multi-dimensional correlation modeling of progress, quality, and environmental constraints. For example, in a certain batch of construction, when the ambient humidity exceeds 75% and the dust concentration is higher than 3.2 mg / m³, the model automatically adjusts the quality risk weights to predict potential issues with poor adhesion of the finishing layer.

[0047] During the backpropagation phase, the physical residual gating unit dynamically labels the update paths of the data-driven branch and the physical constraint branch by calculating the relationship between the gradient direction and the residual magnitude. Over 50 training epochs, the average gradient direction deviation angle of the gating unit was controlled within 15°, resulting in an approximately 26% improvement in training convergence speed and an 18% improvement in stability compared to a model without gating.

[0048] The predicted construction progress, quality risk indicators, and physical constraints output by the model are written into a parameter mapping table and input into the GANomaly reconstruction and anomaly detection module. This module compares the predicted values ​​with the measured data to perform reconstruction, generating a reconstruction residual distribution and anomaly score vector. In a construction monitoring session, the system detected an anomaly score of 0.86 for node 28 (threshold 0.75), automatically generating a residual signal and feeding it back to the parameter update interface between the PIML model and the digital twin model. Based on this, the model adjusted the progress prediction parameters of the data-driven branch and the response coefficients of the physical constraint branch. The progress deviation of the virtual model node decreased from 7.3% to 1.5%, and the quality risk indicator decreased by 32%. Specific experimental data are shown in Table 1. Table 1. Data Table of Digital Twin Intelligent Management and Control for the Entire Process of Interior Decoration Construction Construction node number Average humidity (%) Dust concentration (mg / m³) Schedule forecast deviation (%) Abnormal score Corrected schedule deviation (%) The percentage decrease in the quality risk index (%) N01 68.3 2.7 4.8 0.52 1.9 28 N08 73.6 3.4 6.2 0.79 2.1 33 N15 71.1 3.1 5.7 0.68 1.6 30 N22 75.4 3.5 7.3 0.86 1.5 32 N28 69.8 2.8 5.1 0.58 2.0 25 N34 74.2 3.2 6.8 0.81 1.9 31 N41 72.5 3.0 5.5 0.64 1.7 29 N47 76.7 3.6 7.0 0.84 1.6 33 N52 70.9 3.1 5.3 0.59 1.8 27 N54 73.8 3.4 6.5 0.77 1.9 30 As shown in Table 1, the intelligent management and control system for the entire process of interior decoration construction based on digital twins proposed in this invention has achieved high-precision correction of construction progress prediction and dynamic control of quality risks in practical engineering applications. The system effectively integrates data-driven and physical constraint characteristics, utilizing the GANomaly anomaly detection mechanism and parameter update feedback to form a closed-loop management system. This enables real-time perception, dynamic prediction, and visualization of construction data, significantly improving the intelligence level of construction management and the efficiency of project execution.

[0049] Throughout the construction period, the system detected and corrected abnormal nodes 42 times, with an average correction cycle of 12 hours. Compared with traditional project management models, the application of this invention reduced construction schedule deviations by an average of 38%, improved the construction quality pass rate by approximately 22%, and reduced construction energy consumption by 9%. This digital twin model generates a 3D construction status diagram in real time through a visualization rendering module. The construction status, material usage, and risk indicators of key nodes in the construction period are presented in dynamic diagram form. Project managers can directly view the real-time construction status of each node on the monitoring terminal, realizing data-driven, full-process visualized management.

[0050] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A fully intelligent management and control system for the entire process of interior decoration construction based on digital twins, characterized in that, include: The data acquisition and virtual-real mapping module is used to establish construction nodes and virtual model nodes based on BIM, collect multi-source data and generate a virtual-real mapping database. The improved PIML prediction module is used to receive data-driven features and physical constraint features, output prediction results and write them to the parameter mapping table; The physically constrained attention fusion layer module is used to generate fused feature representations and physically constrained residual signals during forward propagation. The physical residual gating unit module is used to mark the update path and write it into the parameter register area based on the gradient direction and residual magnitude during backpropagation. The GANomaly reconstruction and anomaly detection module is used to reconstruct the prediction results and measured data, and generate the reconstruction feature vector and reconstruction residual distribution. An anomaly scoring and residual signal feedback module is used to generate an anomaly scoring vector from the reconstructed residual distribution, compare thresholds, and form a residual signal. The parameter update interface module is used to send and receive parameters and overwrite virtual model node parameters; The 3D visualization module is used to generate 3D visualization results.

2. The intelligent management and control system for the entire process of interior decoration construction based on digital twins as described in claim 1, characterized in that, The modules are connected in the following way: Step 1: Construct a digital twin model of the interior decoration construction scene, collect multi-source data, preprocess the multi-source data, and establish a virtual-real mapping database; Step 2: Input the feature set extracted from the virtual-real mapping database into the improved PIML prediction model, set up a physical constraint attention fusion layer and a physical residual gating unit. In the forward propagation stage, the physical constraint attention fusion layer generates fused feature representations based on each physical constraint residual. In the back propagation stage, the physical residual gating unit adjusts the update path of the data branch and the physical branch according to the gradient direction and the residual magnitude. Step 3: Perform joint training and inference on the feature set, output the prediction results, and write the prediction results into the parameter mapping table of the digital twin model; Step 4: Input the prediction results and the actual measured data at the construction site into the GANomaly model. By reconstructing the prediction results and the measured data, the reconstructed feature vector and the reconstructed residual distribution are obtained. Step 5: Calculate the anomaly score corresponding to the reconstructed residual distribution, determine the degree of deviation based on the preset threshold, and generate a residual signal when the anomaly score exceeds the threshold. Feedback is sent to the parameter update interface of the improved PIML prediction model and the digital twin model to synchronously correct the physical constraint parameters and virtual model node parameters. Step 6: Update the initial state set and timing information in the corrected digital twin model to generate a 3D visualization result.

3. The intelligent management and control system for the entire process of interior decoration construction based on digital twins as described in claim 2, characterized in that, Step one specifically includes: Construct a digital twin model of the interior decoration construction scenario, and define construction units and divide construction nodes based on the BIM model; Assign a unique identifier and spatial coordinates to each construction node, establish construction dependencies between nodes in the digital twin model, and generate a corresponding virtual model node for each construction node in the digital twin model; Collect multi-source data from BIM models, environmental sensors, construction logs, and video surveillance, and perform timestamp alignment and feature standardization processing on the multi-source data; Based on data attributes, the preprocessed multi-source data is divided into data-driven features and physical constraint features; An initial state set containing data-driven features and physical constraint features is generated for each construction node, and a data mapping relationship is established using the construction node index as the association key. Synchronously record the timing information of construction nodes, including the start time of construction, the planned completion time, and the order of dependencies; Construct a virtual-physical mapping database that includes an initial state set, data-driven features, physical constraint features, construction node indexes, and time sequence information.

4. The intelligent management and control system for the entire process of interior decoration construction based on digital twins as described in claim 2, characterized in that, The physical constraint attention fusion layer generates a fusion feature representation based on the residuals of each physical constraint during the forward propagation phase, specifically as follows: Read the feature set corresponding to each construction node from the virtual-real mapping database, and divide the channel according to data-driven features and physical constraint features; Input the data-driven features into the data-driven branch of the improved PIML prediction model, and input the physical constraint features into the physical constraint branch of the improved PIML prediction model to generate the corresponding branch feature representations. During the forward propagation phase, the physically constrained attention fusion layer is invoked; Perform linear mapping on data-driven branch features to generate query vectors, and perform linear mapping on physical constraint branch features to generate key vectors and value vectors; The attention weight matrix is ​​calculated based on the similarity between the query vector and the key vector, and then normalized. The normalized attention weight matrix and the value vector are weighted and calculated to generate a fused feature representation; The fusion feature representation is matched with the index table corresponding to the construction node identifier, and written into the fusion result buffer in the order of construction nodes; Complete the forward propagation calculation for the current batch, calculate the difference between the physical constraint prediction value output by the physical constraint branch and the result of the preset physical equation, and generate the physical constraint residual signal as an intermediate signal.

5. The intelligent management and control system for the entire process of interior decoration construction based on digital twins as described in claim 2, characterized in that, The physical residual gating unit adjusts the update paths of the data branch and the physical branch according to the gradient direction and residual magnitude during the backpropagation phase, specifically as follows: During the backpropagation phase, the physical residual gating unit is invoked; Perform angle calculation on the gradient directions of the data-driven branch and the physical constraint branch to generate directional relationship identifiers; Read the physical constraint residual signal, perform a threshold comparison operation on the residual amplitude, and generate a residual amplitude identifier; The residual amplitude is the magnitude of the physically constrained residual signal; The gating control unit generates a gating status signal based on the direction relationship identifier and the residual magnitude identifier, and marks the update path of the data-driven branch and the physical constraint branch; After all construction node indexes have been processed, the updated path results are written to the parameter register of the improved PIML prediction model to complete the backpropagation calculation for the current batch.

6. The intelligent management and control system for the entire process of interior decoration construction based on digital twins as described in claim 2, characterized in that, Step three specifically involves: Set the joint training round index and batch size; In the training path, the fused feature representation is fed into the backbone of the improved PIML prediction model to generate data-driven branch output and physical constraint branch output respectively. Construct a training objective consisting of a data prediction error term and a physical consistency constraint term, perform parameter updates, and record the training round index; During the parameter update process, the parameter update interface of the improved PIML prediction model is called, and the updated model parameters are written into the parameter update interface of the digital twin model. Under the inference path, the fusion feature representation is forward-computed based on the latest parameters, and the prediction results are output, including the predicted value of construction progress, quality risk index and physical constraint. The prediction results are written into a parameter mapping table according to the construction node index and time index; Synchronize new records in the parameter mapping table to the virtual-real mapping database, and update the current snapshot of the initial state set and timing information corresponding to the construction node.

7. The intelligent management and control system for the entire process of interior decoration construction based on digital twins as described in claim 2, characterized in that, Step four specifically involves: Read the actual construction site data corresponding to the construction node index and time index from the virtual-real mapping database. The actual construction site data includes progress completion rate, quality inspection parameters and physical response parameters. The prediction results are aligned with the actual measured data at the construction site, and the channels are spliced ​​to generate the input tensor of the GANomaly model and establish a one-to-one correspondence with the construction node index. The encoder module of the GANomaly model is invoked to generate a latent representation vector from the input tensor and write it to the reconstruction buffer. The decoder module of the GANomaly model is invoked to perform reconstruction operations on the latent representation vector and output the reconstructed output vector. Perform element-wise difference operations between the input tensor and the reconstructed output vector to generate the reconstructed feature vector and the reconstructed residual vector; The reconstruction residual vectors are grouped and statistically analyzed according to the construction node index and time index to generate the reconstruction residual distribution; The reconstructed feature vector and the reconstructed residual distribution are written into the anomaly detection result buffer, and an index mapping is established based on the construction node index.

8. The intelligent management and control system for the entire process of interior decoration construction based on digital twins as described in claim 2, characterized in that, Step five specifically involves: Anomaly score vectors are generated by reconstructing the residual distribution; Perform a threshold comparison operation on the abnormal score vectors based on the preset score threshold to generate a deviation status indicator; When the deviation status is marked as exceeding the limit, a residual signal containing the construction node index, time index, anomaly score vector and reconstruction residual distribution is constructed. The residual signal is written into the update queue of the parameter update interface of the improved PIML prediction model and the parameter update interface of the digital twin model. The parameter group is located based on the residual signal, and parameter overwrite is performed in the parameter register area of ​​the improved PIML prediction model. The construction nodes are located based on the residual signals, and the node parameters of the virtual model are overwritten. Write the abnormal scoring vector, deviation status identifier, residual signal and parameter update results of this batch into the correction log area, and synchronize the updated virtual model node parameters to the virtual-real mapping database.

9. The intelligent management and control system for the entire process of interior decoration construction based on digital twins as described in claim 2, characterized in that, Step six specifically includes: The updated virtual model node parameters, initial state set, and timing information are read from the parameter update interface of the digital twin model and the virtual-real mapping database, and a batch of state snapshots are generated according to the construction node index and time index. Write the virtual model node parameters into the node data area of ​​the digital twin model to update the timing information of the construction node; Write the updated initial state set and time series information into the virtual-real mapping database, and append the current snapshot record according to the time index; Load the geometric data and virtual model node set of the BIM model, and establish an index mapping table from the construction node index to the three-dimensional scene entity; Based on the parameter mapping table and virtual model node parameters, the rendering attribute set and annotation information are assigned to the 3D scene entities, and the rendering attribute set is bound to the construction node index using key-value pairs. Perform frame-level rendering on 3D scene entities according to time index, and generate a sequence of 3D visualization results corresponding to the time index; Write the 3D visualization result sequence, index mapping table, and current batch status snapshot to the visualization result storage area, and write the recorded timestamps to the display log area to complete the 3D visualization result output.

Images