Intelligent operation and maintenance system and method for ancient building inheritance and protection based on digital twinning

Abstract

The application belongs to the field of ancient building protection, and provides an ancient building inheritance protection intelligent operation and maintenance system and method based on digital twinning, comprising a multi-source fusion perception and representation module, an ancient building digital twinning body module, an intelligent diagnosis and prediction and early warning module, a multi-target collaborative decision optimization module and a strategy execution and interactive presentation module; the application realizes holographic perception of the state of ancient buildings, forward-looking early warning of risks and collaborative optimization of operation strategies by constructing a closed-loop intelligent operation from monitoring to early warning, early warning to diagnosis, diagnosis to decision-making and decision-making to execution, promotes the change of ancient building protection from experience-driven and passive response to data-driven and active intervention, and provides reliable technical support for the sustainable inheritance of cultural heritage.

Description

Technical Field

[0001] This invention belongs to the field of ancient building protection, specifically an intelligent operation and maintenance system and method for the inheritance and protection of ancient buildings based on digital twins. Background Technology

[0002] Ancient buildings, as irreplaceable cultural heritage, face multiple challenges in their protection and inheritance, including natural structural aging, environmental erosion, human activities, and disaster risks. Traditional protection methods mainly rely on regular manual inspections, simple sensor threshold alarms, and experience-based decision-making, which have the following obvious shortcomings. Existing monitoring systems are mostly focused on a single dimension and lack deep integration of multi-source heterogeneous data, making it difficult to fully reflect the true state of ancient buildings and thus forming data silos. Furthermore, traditional methods are mostly passive responses based on threshold alarms, which cannot perform early diagnosis, evolution prediction, and early warning of damage and risks. Intervention measures are often taken only after problems have become apparent, resulting in high protection costs and limited effectiveness.

[0003] To address these issues, those skilled in the art have proposed an intelligent operation and maintenance system and method for the preservation and protection of ancient buildings based on digital twins. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an intelligent operation and maintenance system and method for the preservation and protection of ancient buildings based on digital twins. This system addresses the shortcomings of existing technologies, where monitoring systems often focus on a single dimension, lack deep integration of multi-source heterogeneous data, and fail to comprehensively reflect the true state of ancient buildings, thus creating data silos. Furthermore, traditional methods are mostly passive responses based on threshold alarms, which cannot perform early diagnosis, evolution prediction, or early warning of damage and risks. Intervention measures are often only taken after problems have become apparent, resulting in high protection costs and limited effectiveness.

[0005] The intelligent operation and maintenance system for the preservation and protection of ancient buildings based on digital twins includes a multi-source fusion perception and representation module, an ancient building digital twin module, an intelligent diagnosis and prediction early warning module, a multi-objective collaborative decision optimization module, and a strategy execution and interactive presentation module. The multi-source fusion sensing and representation module is used to collect and fuse multi-source heterogeneous data from structural health monitoring sensors, microenvironment sensors, video surveillance systems, UAV inspection data and historical repair records. Through feature decoupling and reconstruction mechanisms, it generates a unified and robust representation of the ancient building's condition. The ancient building digital twin module is connected to the multi-source fusion perception and representation module. Based on real-time perception data and historical data, the ancient building digital twin module constructs and dynamically updates a high-precision three-dimensional geometric model, structural mechanics model, material degradation model and environmental field model of the ancient building, forming a high-fidelity virtual mapping that evolves synchronously with the physical entity. The intelligent diagnosis and prediction early warning module is connected to the digital twin module. The intelligent diagnosis and prediction early warning module integrates physical mechanism model and data-driven model for early diagnosis, evolution prediction and graded early warning of structural damage, material aging and environmental risks. The multi-objective collaborative decision optimization module is connected to the digital twin module and the intelligent diagnosis and prediction early warning module. The multi-objective collaborative decision optimization module has a built-in decision framework based on multi-agent reinforcement learning, which is used to simulate and optimize the conflict between protection priority, visitor experience and operation safety objectives in the digital twin environment, and output Pareto optimal operation and maintenance strategy. The strategy execution and interactive presentation module is used to distribute the optimized inspection plan, tourist flow guidance scheme, environmental control instructions, and emergency response plan to the relevant execution terminals, and to provide an immersive interactive interface to managers through a 3D visualization platform.

[0006] Furthermore, the multi-source fusion perception and representation module specifically includes: a spatiotemporal alignment and data quality control unit, a feature decoupling variational autoencoder unit, and a feature completion unit; The spatiotemporal alignment and data quality control unit is used to perform time synchronization and spatial registration of multi-source heterogeneous data, and adopts a hybrid model based on sliding window statistics and isolated forest algorithm to detect and repair abnormal data. The encoder network of the feature decoupled variational autoencoder unit maps the quality-controlled multi-source data to a low-dimensional latent space, and by introducing a decoupling regularization term, the different dimensions in the latent space independently correspond to at least four physical factors among structural vibration, temperature and humidity, crowd density, surface deformation, light intensity, and pollutant concentration. The feature completion unit is connected to the feature decoupling variational autoencoder unit, and is used to reconstruct the complete feature vector based on the remaining dimensions by omitting the corresponding latent variable dimension when a specific sensor signal is missing.

[0007] Furthermore, the training objective function of the feature decoupling variational autoencoder unit is: ; in To reconstruct the loss term, For the latent variable distribution and the standard normal prior Divergence, determined by hyperparameters Control its intensity, To decouple the loss term, an approach based on total correlation is adopted to minimize the statistical dependencies between the dimensions of the latent variables, determined by hyperparameters. Control its intensity; The decoupling loss term The following formula is used for calculation: ; in, , The i and j-th dimensions of the latent variables are represented, and corr represents the correlation coefficient calculated on the batch data.

[0008] Furthermore, the ancient building digital twin module adopts BIM+GIS fusion modeling technology, integrating finite element analysis model, computational fluid dynamics model, material degradation dynamics model and visualization rendering engine; The finite element analysis model is used to simulate the stress, strain and displacement response of the structure under wind, rain, snow, earthquake and crowd loads. The computational fluid dynamics model is used to simulate indoor and outdoor temperature and humidity distribution, air flow, and pollutant diffusion. The material degradation kinetics model predicts the degradation rate of wood, brick, and mural materials based on temperature, humidity, light, and pollutant concentration. The visualization rendering engine supports real-time lighting and shadow rendering, highlighting of damaged areas, and historical state retrospection and comparison.

[0009] Furthermore, the intelligent diagnosis and prediction early warning module includes a physically guided deep neural network unit, a spatiotemporal graph attention network, and an early warning classification and push unit; The training loss function of the physically guided deep neural network unit fuses data prediction error and physical mechanism residuals: ; in, and Here, represents the weighting coefficients, controlling the impact of data loss and physical loss respectively; represents the parameters in the physical equations. This is a data-driven loss term used to measure the model's predicted values. and actual observed values The differences between them mean that different loss functions can be used for different tasks. The physical consistency loss term forces the model prediction results to conform to known physical laws. This loss is constructed based on the physical residual, which is the imbalance in the physical equation after substituting the predicted value.

[0010] Furthermore, the multi-objective collaborative decision optimization module defines a structural safety intelligent agent, a tourist experience intelligent agent, and an operating cost intelligent agent; The utility function of the structural safety intelligent agent is the negative value of the cumulative structural damage risk, and the risk is assessed in real time through the finite element model in the digital twin; The utility function of the visitor experience agent is calculated based on a comprehensive assessment of visitor density comfort, visitor path smoothness, and visual viewing integrity. The utility function of the operating cost agent is a negative weighted sum of energy consumption, maintenance costs, and labor costs; Each agent approaches the Nash equilibrium through multiple rounds of iterative game in a digital twin environment and outputs an equilibrium strategy.

[0011] Furthermore, the game-theoretic iteration process of the multi-objective collaborative decision-making optimization module includes the following steps: Step 1: The agent initializes its strategy based on the current digital twin state; Step 2: In each iteration, each agent, assuming that the policies of other agents remain unchanged, searches for the optimal response policy that maximizes its own utility through simulation. Step 3: Update the strategy and repeat until the strategy change is less than the threshold, then output the balanced strategy combination.

[0012] 8. The intelligent operation and maintenance system for the preservation and protection of ancient buildings based on digital twins as described in claim 1, characterized in that: the system adopts a hierarchical federated learning architecture for model evolution and data privacy protection, specifically employing the following steps: Step 1: Deploy a global model in the cloud, aggregating model updates from multiple ancient building sites; Step 2: Deploy edge nodes locally on each ancient building, run the local model, and train it based on private data; Step 3: In the feature decoupling variational autoencoder, the latent variable dimensions related to general physical laws participate in federated aggregation, while the dimensions related to building-specific attributes are only updated locally. Step 4: Differential privacy is used to protect the privacy of model parameters before uploading.

[0013] A smart operation and maintenance method for the preservation and protection of ancient buildings based on digital twins includes the following steps: Step 1: Acquire real-time data on the structure, environment, and human activities of ancient buildings using multi-source sensors and acquisition equipment; Step 2: Perform spatiotemporal alignment, anomaly detection, and feature decoupling and fusion on multi-source data to generate a unified state representation; Step 3: Dynamically update the digital twin of ancient buildings based on fused data; Step 4: Use physics-guided deep learning models to diagnose and predict structural health, material aging, and environmental risks. Step 5: In the digital twin environment, the protection, experience, and cost objectives are collaboratively optimized through multi-agent game theory to generate the optimal operation and maintenance strategy; Step 6: Decode the strategy into specific execution instructions and issue them, while displaying and interacting with the status through a visualization platform.

[0014] Furthermore, the feature decoupling fusion in step two is performed using a variational autoencoder, which introduces a decoupling loss term during its training process, so that the dimensions of the latent variables correspond independently to the physical factors.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention utilizes a feature decoupling variational autoencoder in the multi-source fusion sensing and representation module to map multi-source heterogeneous data from structural health monitoring, microenvironment sensing, video surveillance, drone inspections, and historical repair records to a decoupled latent space. A decoupling loss term is introduced during training, allowing each dimension of the latent variables to independently correspond to physical factors such as structural vibration, temperature and humidity, pedestrian density, and surface deformation. This achieves deep feature fusion and physical interpretability, effectively breaking down data silos and improving the robustness and completeness of state representation. When some sensor signals are missing, the system can empty the corresponding dimensions in the latent space and reconstruct the complete feature vector through the decoder, ensuring the system's continuous and stable operation even with incomplete data.

[0016] 2. This invention constructs a high-fidelity digital twin of ancient buildings that evolves synchronously with the physical entity. It integrates BIM+GIS fusion modeling, finite element analysis, computational fluid dynamics, material degradation dynamics, and a visualization rendering engine to achieve multi-physics field coupled simulation and real-time visualization of structural mechanical response, environmental field distribution, and material degradation process. It supports highlighting of damaged areas, historical state retrospection, and immersive interaction, providing a high-fidelity and retrospective virtual simulation environment for protection decisions, and significantly improving the scientific nature and foresight of operation and maintenance decisions.

[0017] 3. This invention deploys a collaborative decision-making optimization module based on multi-agent reinforcement learning in a digital twin environment. It models the multi-objective conflict between structural safety, visitor experience, and operating costs as the utility function of virtual agents. Through multi-round game simulation, it autonomously approximates the Nash equilibrium and outputs the Pareto optimal operation and maintenance strategy. This achieves closed-loop intelligent operation and maintenance from monitoring to early warning, early warning to diagnosis, diagnosis to decision-making, and decision-making to execution, thereby improving the systematicness, coordination, and economy of ancient building protection and reducing reliance on human experience. Attached Figure Description

[0018] Figure 1 This is a block diagram of the overall system structure of the present invention; Figure 2 This is a system architecture diagram of the multi-source fusion sensing and representation module in this invention; Figure 3 This is a system architecture diagram of the digital twin module for ancient architecture in this invention; Figure 4 This is a system architecture diagram of the intelligent diagnosis and prediction early warning module in this invention; Figure 5This is a system architecture diagram of the multi-objective collaborative decision optimization module in this invention. Detailed Implementation

[0019] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0020] As attached Figure 1 To be continued Figure 5 As shown: Example 1: This invention provides an intelligent operation and maintenance system and method for the preservation and protection of ancient buildings based on digital twins, including a multi-source fusion perception and representation module, an ancient building digital twin module, an intelligent diagnosis and prediction early warning module, a multi-objective collaborative decision optimization module, and a strategy execution and interactive presentation module; The multi-source fusion sensing and representation module is used to collect and fuse multi-source heterogeneous data from structural health monitoring sensors, microenvironment sensors, video surveillance systems, UAV inspection data and historical repair records. Through feature decoupling and reconstruction mechanisms, it generates a unified and robust representation of the ancient building's condition. The multi-source fusion sensing and representation module is deployed at the edge node of the ancient building site, and the hardware includes: Sensor network: Install vibration acceleration sensors and strain sensors in key structural parts (such as beams, columns, and mortise and tenon joints); deploy environmental sensors such as temperature, humidity, light, CO2, VOC, and PM2.5 indoors and outdoors; install video cameras and people counting equipment along the main visitor routes; Inspection equipment: Regularly use drones equipped with high-definition cameras and laser scanners to acquire high-precision images and point cloud data of building facades and roofs; Data access layer: Sensor data is uploaded to the edge server in real time via wireless transmission protocols such as LoRa, NB-IoT, and 5G; The ancient building digital twin module is connected to the multi-source fusion perception and representation module. Based on real-time perception data and historical data, the ancient building digital twin module constructs and dynamically updates a high-precision three-dimensional geometric model, structural mechanics model, material degradation model and environmental field model of the ancient building, forming a high-fidelity virtual mapping that evolves synchronously with the physical entity. The intelligent diagnosis and prediction early warning module is connected to the digital twin module. The intelligent diagnosis and prediction early warning module integrates physical mechanism model and data-driven model, and is used for early diagnosis, evolution prediction and graded early warning of various abnormal states such as structural damage, material aging and environmental risks. The multi-objective collaborative decision optimization module is connected to the digital twin module and the intelligent diagnosis and prediction early warning module. The multi-objective collaborative decision optimization module has a built-in decision framework based on multi-agent reinforcement learning, which is used to simulate and optimize the conflict of multiple objectives such as protection priority, tourist experience, and operational safety in the digital twin environment, and output the Pareto optimal operation and maintenance strategy. The strategy execution and interactive presentation module is used to distribute the optimized inspection plan, tourist flow guidance scheme, environmental control instructions, and emergency response plan to the relevant execution terminals, and to provide an immersive interactive interface to managers through a 3D visualization platform.

[0021] It should be further explained in this embodiment that the multi-source fusion perception and representation module specifically includes: a spatiotemporal alignment and data quality control unit, a feature decoupling variational autoencoder unit, and a feature completion unit; The spatiotemporal alignment and data quality control unit is used to perform time synchronization and spatial registration of multi-source heterogeneous data, and adopts a hybrid model based on sliding window statistics and isolated forest algorithm to detect and repair abnormal data. Spatiotemporal alignment specifically involves unifying the timestamps of all sensor data to UTC time and spatially mapping them using GIS coordinates and component IDs in the BIM model. The anomaly detection uses the sliding window Z-score algorithm to detect mutation data, combined with the isolated forest algorithm to identify long-term drift anomalies, and uses spatiotemporal KNN interpolation for missing values. The encoder network of the Feature Decoupling Variational Autoencoder (FD-VAE) unit maps quality-controlled multi-source data to a low-dimensional latent space. By introducing a decoupling regularization term, different dimensions in the latent space independently correspond to at least four physical or anthropogenic factors from structural vibration, temperature and humidity, pedestrian density, surface deformation, light intensity, and pollutant concentration. The FD-VAE is used, with its encoder being a three-layer fully connected network (input dimension dynamically adjusted according to the number of sensors, output latent space dimension set to 16), and a symmetrical decoder structure. During training, the network is configured... =0.5, =1.2, using Adam optimizer, learning rate 0.001, batch size 64; The feature completion unit is connected to the feature decoupling variational autoencoder unit. When a specific sensor signal is missing, it can reconstruct the complete feature vector based on the remaining dimensions by setting the corresponding latent variable dimension to null and using the decoder network, thereby enabling the system to continue operating even when some data is missing.

[0022] In this embodiment, it should be further explained that the training objective function of the feature decoupling variational autoencoder unit is: ; in To reconstruct the loss term, For the latent variable distribution and the standard normal prior Divergence, determined by hyperparameters Control its intensity, To decouple the loss term, an approach based on total correlation is adopted to minimize the statistical dependencies between the dimensions of the latent variables, determined by hyperparameters. Control its intensity; The decoupling loss term The following formula is used for calculation: ; in, , The i and j-th dimensions of the latent variables are represented, and corr represents the correlation coefficient calculated on the batch data.

[0023] It should be further explained in this embodiment that the ancient building digital twin module adopts BIM+GIS fusion modeling technology, integrating finite element analysis model, computational fluid dynamics model, material degradation dynamics model and visualization rendering engine; The finite element analysis model is used to simulate the stress, strain and displacement response of the structure under wind, rain, snow, earthquake and crowd loads; specifically, a finite element model is established in ANSYS, the material parameters are set according to measured data (such as the elastic modulus of wood and the compressive strength of bricks and stones), and the load conditions are dynamically updated according to real-time meteorological data and crowd flow data. The computational fluid dynamics model is used to simulate indoor and outdoor temperature and humidity distribution, air flow, and pollutant diffusion; specifically, Fluent is used for CFD simulation, and boundary conditions are provided in real time by sensors to simulate indoor and outdoor temperature and humidity, airflow organization, and pollutant diffusion.

[0024] The material degradation kinetics model predicts the degradation rate of materials such as wood, brick, and murals based on temperature, humidity, light, and pollutant concentration; specifically, it predicts the remaining lifespan of different materials based on the Arrhenius equation and the light aging model, combined with real-time environmental data. The visualization rendering engine supports real-time lighting and shadow rendering, highlighting of damaged areas, and historical state retrospection and comparison. The visualization rendering engine uses Unity3D + digital twin platform (such as Tencent Cloud TDS, Huawei Cloud ModelArts) to achieve lightweight model and WebGL publishing, and supports VR / AR immersive browsing.

[0025] This embodiment requires further explanation. The intelligent diagnosis and prediction early warning module includes a physically guided deep neural network unit, a spatiotemporal graph attention network, and an early warning classification and push unit. The intelligent diagnosis and prediction early warning module is deployed on a cloud-based AI platform. The physically guided deep neural network unit uses ResNet-50 as its backbone, with multi-source fusion features as input and damage probability, displacement prediction value, etc., as output. The physical loss term is constructed based on structural dynamics equations (such as motion equations), with weights... =0.7, =0.3; The spatiotemporal graph attention network treats building components as graph nodes and sensor data as node features to construct a spatiotemporal graph model for predicting damage propagation paths; The early warning classification and push unit is divided into four levels according to risk level: blue (low), yellow (medium), orange (high), and red (emergency), and pushes the warning through multiple channels such as SMS, APP, and large screen.

[0026] The training loss function of the physically guided deep neural network unit fuses data prediction error and physical mechanism residuals: ; in, and Here, represents the weighting coefficients, controlling the impact of data loss and physical loss respectively; represents the parameters in the physical equations. This is a data-driven loss term; used to measure the model's predicted values. and actual observed values The differences between them mean that different loss functions can be used for different tasks: For regression tasks (such as displacement prediction): mean squared error is used. ; in This refers to the number of time steps or the number of samples. and For the first Predicted and actual values ​​at each time step; For classification tasks (such as damage recognition): cross-entropy loss is used; The physical consistency loss term forces the model predictions to conform to known physical laws. This loss is constructed based on physical residuals, i.e., the unbalanced quantities in the physical equations after substituting the predicted values. Taking structural mechanics as an example, if the simplified Navier-Stokes equations are used to describe the wind-induced vibrations of ancient buildings, then: ; in For the velocity vector field, For time, For gradient operators, For fluid density, For pressure field, The dynamic viscosity coefficient, The norm, usually the L2 norm, represents the magnitude of the residual.

[0027] In this embodiment, it should be further explained that the multi-objective collaborative decision optimization module defines a structural safety agent, a tourist experience agent, and an operating cost agent; the multi-objective collaborative decision optimization module adopts a multi-agent reinforcement learning (MARL) framework and is implemented based on Python + RayRLlib; The utility function of the structural safety agent is the negative value of the accumulated structural damage risk, and the risk is assessed in real time through the finite element model in the digital twin; the state of the structural safety agent includes the component stress ratio, vibration frequency, and material degradation index; the action is to suggest reinforcement locations and methods; and the reward function is the percentage reduction in risk. The utility function of the visitor experience agent is calculated based on a comprehensive assessment of visitor density comfort, visitor path smoothness, and visual viewing integrity; the status of the visitor experience agent includes the population density, dwell time, and temperature and humidity comfort in each area; the actions include adjusting the flow path and controlling the air conditioning; and the reward function is the visitor satisfaction questionnaire score. The utility function of the operating cost agent is the negative weighted sum of energy consumption, maintenance costs, and labor costs; the state of the operating cost agent is energy consumption, maintenance costs, and manpower allocation; the actions are equipment start-up and shutdown, and inspection frequency adjustment; the reward function is the cost saving ratio. Each agent approaches a Nash equilibrium through multiple rounds of iterative game in a digital twin environment, outputting an equilibrium policy. Specifically, the Nash Q-learning algorithm is used, with an upper limit of 500 iterations and a policy convergence threshold. In a digital twin environment, multiple strategy combinations are simulated in parallel to output a Pareto optimal solution set; The game-theoretic iterative process of the multi-objective collaborative decision-making optimization module includes the following steps: Step 1: The agent initializes its strategy based on the current digital twin state; Step 2: In each iteration, each agent, assuming that the policies of other agents remain unchanged, searches for the optimal response policy that maximizes its own utility through simulation. Step 3: Update the strategy and repeat until the strategy change is less than the threshold or the maximum number of iterations is reached, and output the balanced strategy combination.

[0028] This embodiment requires further explanation that the system employs a hierarchical federated learning architecture for model evolution and data privacy protection, specifically through the following steps: Step 1: Deploy a global model in the cloud, aggregating model updates from multiple ancient building sites; Step 2: Deploy edge nodes locally on each ancient building, run the local model, and train it based on private data; Step 3: In the feature decoupling variational autoencoder, the latent variable dimensions related to general physical laws participate in federated aggregation, while the dimensions related to building-specific attributes are only updated locally; specifically, the cloud aggregates model updates uploaded by each site every 24 hours (only general feature dimensions are aggregated), using the FedAvg algorithm; Step 4: Before uploading the model parameters, differential privacy or homomorphic encryption technology is used for privacy protection; specifically, Gaussian noise is added to the parameters before uploading. Alternatively, Paillier homomorphic encryption can be used.

[0029] In this embodiment, it should be further explained that the strategy execution and interactive presentation module converts the optimized strategy into JSON instructions and sends them to the field execution terminal, such as smart meters, guide screens, inspection robots, and air conditioning systems, via the MQTT protocol. It also develops a Web 3D visualization interface based on Vue + Three.js. The 3D visualization platform in the strategy execution and interactive presentation module supports real-time data dashboards, highlighting of damaged areas and historical comparisons, timelines of early warning events, strategy simulation and deduction, and VR inspection mode.

[0030] As can be seen from the above, this embodiment, through a multi-source fusion sensing and representation module deployed on the ancient building site, transforms heterogeneous data from multiple sources such as vibration, temperature and humidity, pedestrian flow, and images into a unified and robust state representation through spatiotemporal alignment, anomaly detection, and feature decoupling and fusion. This representation not only has clear physical meaning (such as independent correspondence between structural vibration and surface deformation), but also maintains continuous system operation through a feature completion mechanism when some sensors are missing, greatly improving the integrity and reliability of ancient building state monitoring. Combined with BIM+GIS fusion modeling and multiphysics simulation, this embodiment constructs a high-fidelity, interactive digital twin of the ancient building, realizing accurate mapping and real-time extrapolation from physical entities to virtual space, providing a reliable simulation environment for subsequent intelligent diagnosis and decision optimization.

[0031] Example 2: A smart operation and maintenance method for the preservation and protection of ancient buildings based on digital twins, comprising the following steps: Step 1: Through the deployed sensor network, video system and drones, collect structural vibration, environmental parameters, crowd density, images and point cloud data in real time. The sampling frequency is set according to the data type (e.g., 100Hz for vibration data, 1Hz for temperature and humidity). Step 2: Perform spatiotemporal alignment and quality control on the multi-source data, input the trained FD-VAE, and output a 16-dimensional decoupled feature vector; if data is missing, activate the feature completion mechanism. Step 3: Inject the fusion features into the BIM+GIS model to drive real-time simulation of finite element, CFD, and material degradation models, and update the 3D visualization scene to render the latest state; Step 4: Use PGDNN to perform real-time analysis on the feature vectors, output damage identification results and risk prediction, and combine the ST-GAT model to perform spatiotemporal evolution analysis to generate early warning information. Step 5: Initiate multi-agent game in the digital twin environment, and simulate the impact of different operation and maintenance strategies on security, experience and cost. Then output the equilibrium strategy combination, such as "flow restriction during high temperature period in summer + enhanced ventilation + inspection of key parts". Step Six: Parse the strategy into specific control instructions and send them to the execution terminal. At the same time, display the strategy execution effect and the real-time status of the building in the visualization platform, supporting interactive adjustments by management personnel.

[0032] The feature decoupling and fusion in step two is performed using a variational autoencoder. During its training, a decoupling loss term is introduced to make the dimensions of the latent variables correspond independently to the physical factors. When some data is missing, feature completion is achieved by setting the latent variables to zero and reconstructing the decoder.

[0033] Working principle: By deploying sensors such as vibration, strain, temperature and humidity, light, and video sensors at key parts of ancient buildings, as well as regular drone inspections, multi-dimensional monitoring data is collected in real time. After spatiotemporal alignment, anomaly detection and repair, all data is input into a feature decoupled variational autoencoder (FD-VAE), mapped to a decoupled latent space, and generates a unified state representation with physical interpretability. If a certain type of data (such as a vibration sensor failure) is detected, the system automatically emptys the corresponding dimension in the latent space and reconstructs the complete feature through a decoder to ensure that downstream processing is not interrupted. Based on the above-mentioned fusion features, the digital twin of ancient buildings is updated in real time. The BIM+GIS model provides geometric and spatial benchmarks, the finite element model calculates the mechanical response of the structure under real-time wind, snow and crowd loads, the CFD model simulates indoor and outdoor temperature and humidity and pollutant diffusion, the material degradation model predicts the deterioration rate of wood, bricks and stones in the current environment, and all simulation results are rendered in real time through the visualization engine. This allows managers to highlight risk areas in the 3D scene, compare historical status, and conduct VR / AR immersive inspections. A physical-guided deep neural network receives physical field data from fused features and digital twin outputs, performing multi-task analysis such as structural damage identification, displacement prediction, and aging assessment. A spatiotemporal graph attention network further models the spatiotemporal propagation relationship of damage between building components. Diagnostic results are divided into four levels—blue, yellow, orange, and red—based on risk severity and are pushed to relevant responsible persons in real time through multiple channels such as SMS, APP, and command screen. In a digital twin environment, a multi-agent game optimization process is initiated. The three agents, structural safety, tourist experience, and operating cost, aim to minimize damage risk, maximize tourist satisfaction, and minimize operation and maintenance costs, respectively. Through multiple rounds of iteration, a Nash equilibrium strategy is sought. This strategy is the best compromise solution for the multi-objective conflict in the current state, such as "limiting the flow during high-temperature periods + strengthening ventilation + increasing patrols in key areas". The optimization strategy is decoded into specific control instructions by the strategy execution and interactive presentation module and sent to the field terminal through protocols such as MQTT. These instructions include adjusting the temperature and humidity of the air conditioner, releasing tourist flow guidance information, and generating robot inspection paths. At the same time, the execution effect of all instructions and the real-time status of the building are displayed synchronously on the 3D visualization platform, which supports managers to dynamically adjust the strategy or initiate manual intervention. The data generated during the system operation continuously flows back to the perception and digital twin module, forming a closed-loop iterative optimization.

[0034] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Any changes, modifications, substitutions and variations made by those skilled in the art to the above embodiments within the scope of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital twin-based intelligent operation and maintenance system for the preservation and protection of ancient buildings, characterized by: It includes a multi-source fusion perception and representation module, an ancient building digital twin module, an intelligent diagnosis and prediction early warning module, a multi-objective collaborative decision optimization module, and a strategy execution and interactive presentation module; The multi-source fusion sensing and representation module is used to collect and fuse multi-source heterogeneous data from structural health monitoring sensors, microenvironment sensors, video surveillance systems, UAV inspection data and historical repair records. Through feature decoupling and reconstruction mechanisms, it generates a unified and robust representation of the ancient building's condition. The ancient building digital twin module is connected to the multi-source fusion perception and representation module. Based on real-time perception data and historical data, the ancient building digital twin module constructs and dynamically updates a high-precision three-dimensional geometric model, structural mechanics model, material degradation model and environmental field model of the ancient building, forming a high-fidelity virtual mapping that evolves synchronously with the physical entity. The intelligent diagnosis and prediction early warning module is connected to the digital twin module. The intelligent diagnosis and prediction early warning module integrates physical mechanism model and data-driven model for early diagnosis, evolution prediction and graded early warning of structural damage, material aging and environmental risks. The multi-objective collaborative decision optimization module is connected to the digital twin module and the intelligent diagnosis and prediction early warning module. The multi-objective collaborative decision optimization module has a built-in decision framework based on multi-agent reinforcement learning, which is used to simulate and optimize the conflict between protection priority, visitor experience and operation safety objectives in the digital twin environment, and output Pareto optimal operation and maintenance strategy. The strategy execution and interactive presentation module is used to distribute the optimized inspection plan, tourist flow guidance scheme, environmental control instructions, and emergency response plan to the relevant execution terminals, and to provide an immersive interactive interface to managers through a 3D visualization platform.

2. The intelligent operation and maintenance system for the preservation and protection of ancient buildings based on digital twins as described in claim 1, characterized in that: The multi-source fusion perception and representation module specifically includes: a spatiotemporal alignment and data quality control unit, a feature decoupling variational autoencoder unit, and a feature completion unit; The spatiotemporal alignment and data quality control unit is used to perform time synchronization and spatial registration of multi-source heterogeneous data, and adopts a hybrid model based on sliding window statistics and isolated forest algorithm to detect and repair abnormal data. The encoder network of the feature decoupled variational autoencoder unit maps the quality-controlled multi-source data to a low-dimensional latent space, and by introducing a decoupling regularization term, the different dimensions in the latent space independently correspond to at least four physical factors among structural vibration, temperature and humidity, crowd density, surface deformation, light intensity, and pollutant concentration. The feature completion unit is connected to the feature decoupling variational autoencoder unit, and is used to reconstruct the complete feature vector based on the remaining dimensions by omitting the corresponding latent variable dimension when a specific sensor signal is missing.

3. The intelligent operation and maintenance system for the preservation and protection of ancient buildings based on digital twins as described in claim 1, characterized in that: The training objective function of the feature decoupling variational autoencoder unit is: ； in To reconstruct the loss term, For the latent variable distribution and the standard normal prior Divergence, determined by hyperparameters Control its intensity, To decouple the loss term, a method based on total correlation is adopted to minimize the statistical dependencies between the dimensions of the latent variables, determined by the hyperparameters. Control its intensity; The decoupling loss term The following formula is used for calculation: ； in, , The i and j-th dimensions of the latent variables are represented, and corr represents the correlation coefficient calculated on the batch data.

4. The intelligent operation and maintenance system for the preservation and protection of ancient buildings based on digital twins as described in claim 1, characterized in that: The ancient building digital twin module adopts BIM+GIS fusion modeling technology, integrating finite element analysis model, computational fluid dynamics model, material degradation dynamics model and visualization rendering engine; The finite element analysis model is used to simulate the stress, strain and displacement response of the structure under wind, rain, snow, earthquake and crowd loads. The computational fluid dynamics model is used to simulate indoor and outdoor temperature and humidity distribution, air flow, and pollutant diffusion. The material degradation kinetics model predicts the degradation rate of wood, brick, and mural materials based on temperature, humidity, light, and pollutant concentration. The visualization rendering engine supports real-time lighting and shadow rendering, highlighting of damaged areas, and historical state retrospection and comparison.

5. The intelligent operation and maintenance system for the preservation and protection of ancient buildings based on digital twins as described in claim 1, characterized in that: The intelligent diagnosis and prediction early warning module includes a physically guided deep neural network unit, a spatiotemporal graph attention network, and an early warning classification and push unit; The training loss function of the physically guided deep neural network unit fuses data prediction error and physical mechanism residuals: ； in, and Here, are weighting coefficients that control the effects of data loss and physical loss, respectively; are parameters in the physical equations. This is a data-driven loss term used to measure the model's predicted values. and actual observed values The differences between them mean that different loss functions can be used for different tasks. The physical consistency loss term forces the model prediction results to conform to known physical laws. This loss is constructed based on the physical residual, which is the imbalance in the physical equation after substituting the predicted value.

6. The intelligent operation and maintenance system for the preservation and protection of ancient buildings based on digital twins as described in claim 1, characterized in that: The multi-objective collaborative decision-making optimization module defines a structural safety intelligent agent, a visitor experience intelligent agent, and an operating cost intelligent agent; The utility function of the structural safety intelligent agent is the negative value of the cumulative structural damage risk, and the risk is assessed in real time through the finite element model in the digital twin; The utility function of the visitor experience agent is calculated based on a comprehensive assessment of visitor density comfort, visitor path smoothness, and visual viewing integrity. The utility function of the operating cost agent is a negative weighted sum of energy consumption, maintenance costs, and labor costs; Each agent approaches the Nash equilibrium through multiple rounds of iterative game in a digital twin environment and outputs an equilibrium strategy.

7. The intelligent operation and maintenance system for the preservation and protection of ancient buildings based on digital twins as described in claim 6, characterized in that: The game-theoretic iterative process of the multi-objective collaborative decision-making optimization module includes the following steps: Step 1: The agent initializes the strategy based on the current digital twin state; Step 2: In each iteration, each agent, assuming that the policies of other agents remain unchanged, searches for the optimal response policy that maximizes its own utility through simulation. Step 3: Update the strategy and repeat until the strategy change is less than the threshold, then output the balanced strategy combination.

8. The intelligent operation and maintenance system for the preservation and protection of ancient buildings based on digital twins as described in claim 1, characterized in that: The system employs a hierarchical federated learning architecture for model evolution and data privacy protection, specifically using the following steps: Step 1: Deploy a global model in the cloud, aggregating model updates from multiple ancient building sites; Step 2: Deploy edge nodes locally on each ancient building, run the local model, and train it based on private data; Step 3: In the feature decoupling variational autoencoder, the latent variable dimensions related to general physical laws participate in federated aggregation, while the dimensions related to building-specific attributes are only updated locally. Step 4: Differential privacy is used to protect the privacy of model parameters before uploading.

9. A digital twin-based intelligent operation and maintenance method for the preservation and protection of ancient buildings, applied to the digital twin-based intelligent operation and maintenance system for the preservation and protection of ancient buildings as described in any one of claims 1-8, characterized in that: Includes the following steps: Step 1: Acquire real-time data on the structure, environment, and human activities of ancient buildings using multi-source sensors and acquisition equipment; Step 2: Perform spatiotemporal alignment, anomaly detection, and feature decoupling and fusion on multi-source data to generate a unified state representation; Step 3: Dynamically update the digital twin of ancient buildings based on fused data; Step 4: Use physics-guided deep learning models to diagnose and predict structural health, material aging, and environmental risks. Step 5: In the digital twin environment, the protection, experience, and cost objectives are collaboratively optimized through multi-agent game theory to generate the optimal operation and maintenance strategy; Step 6: Decode the strategy into specific execution instructions and issue them, while displaying and interacting with the status through a visualization platform.

10. The intelligent operation and maintenance method for the preservation and inheritance of ancient buildings based on digital twins as described in claim 9, characterized in that: The feature decoupling and fusion in step two is performed using a variational autoencoder, which introduces a decoupling loss term during its training process to make the dimensions of latent variables correspond independently to the physical factors.

Images