Building engineering safety monitoring and dynamic evaluation method based on BIM technology

Abstract

The invention provides a building engineering safety monitoring and dynamic evaluation method based on a BIM technology, and relates to the technical field of building engineering, and the method comprises the following steps: collecting the monitoring data of a building structure through a multi-source sensor, including environment parameters and building parameters; a multi-model fusion prediction framework is adopted to dynamically predict and evaluate the structural health condition of a building, and the multi-model fusion prediction framework comprises the following parts: an improved multi-scale long-short-term memory network module used for capturing short-term and long-term dependence information in building structure monitoring data; the self-adaptive time-varying Bayesian network module is used for dynamically adjusting the conditional probability in risk assessment according to the change of the environmental condition and the building load; the random forest regression module is used for processing nonlinear high-dimensional data in building monitoring and performing multi-dimensional feature fusion; the emergency response scheduling module is used for handling emergencies; and obtaining a final building safety state prediction result through weighted fusion.

Description

Technical Field

[0001] The present invention relates to the technical field of construction engineering, and in particular to a construction engineering safety monitoring and dynamic evaluation method based on BIM technology. Background Art

[0002] According to the safety risk assessment method in the construction of existing buildings disclosed in the Chinese patent publication number "CN110991918B", a finite element model of the existing building structure is established; the main structure of the existing building is divided according to the finite element model; the main structure includes columns, beams and trusses; the safety risk of the existing building is assessed by the load strength reserve ratio of the main structure. This assessment method can provide risk warning for the structural safety of existing buildings in the construction project.

[0003] According to the construction project quality monitoring system based on BIM technology disclosed in China Patent Publication No. "CN116823064A", it includes a server, a BIM model construction module, a building deviation monitoring and management module, a construction material inspection and supervision module and a construction environmental impact assessment module; the present invention performs building deviation comparison detection on the corresponding construction project to determine whether the building deviation of the corresponding constructed part of the construction project is qualified, and inspects and analyzes all construction materials of the corresponding construction project one by one to determine whether the quality of the corresponding construction materials is qualified, and analyzes the environmental conditions of all sub-areas of the corresponding construction project one by one to determine whether the environment of the corresponding sub-area is qualified, and sets a comprehensive assessment cycle and conducts project supervision intensity assessment and analysis.

[0004] The above patent documents and prior art have the following technical problems when used:

[0005] Traditional building health status prediction mostly relies on a single model, which is easily affected by data fluctuations or anomalies, resulting in inaccurate prediction results;

[0006] Traditional emergency response scheduling is usually static and cannot quickly adjust resource allocation according to the real-time status of the building, resulting in slow emergency response and low resource utilization. Summary of the invention

[0007] Technical issues solved:

[0008] In view of the shortcomings of the prior art, the present invention provides a construction engineering safety monitoring and dynamic assessment method based on BIM technology, which solves the following problems:

[0009] 1. Traditional building health status prediction mostly relies on a single model, which is easily affected by data fluctuations or anomalies, resulting in inaccurate prediction results;

[0010] 2. Traditional emergency response scheduling is usually static and cannot quickly adjust resource allocation according to the real-time status of the building, resulting in slow emergency response and low resource utilization.

[0011] Technical solution:

[0012] To achieve the above objectives, the present invention is implemented through the following technical solutions: a construction engineering safety monitoring and dynamic assessment method based on BIM technology, the method comprising the following steps:

[0013] Sp1, collect monitoring data of building structures through multi-source sensors, including environmental parameters and building parameters;

[0014] Sp2. Use a multi-model fusion prediction framework to dynamically predict and evaluate the structural health status of the building. The multi-model fusion prediction framework includes the following parts:

[0015] Sp2.1, improved multi-scale long short-term memory network module to capture short-term and long-term dependency information in building structure monitoring data;

[0016] Sp2.2, Adaptive Time-Varying Bayesian Network Module, used to dynamically adjust the conditional probability in risk assessment according to changes in environmental conditions and building loads;

[0017] Sp2.3, random forest regression module, is used to process nonlinear high-dimensional data in building monitoring and perform multi-dimensional feature fusion;

[0018] Sp3, emergency response dispatch module, used to handle emergency situations;

[0019] Sp4, based on the data output of Sp1-Sp3, obtains the final building safety status prediction result through weighted fusion.

[0020] Preferably, the improved multi-scale long short-term memory network module adopts an improved multi-scale long short-term memory network model, and the improved multi-scale long short-term memory network model is as follows:

[0021] : Output features of the i-th convolution kernel; : The weight matrix of the i-th convolution kernel, representing the parameters used to extract specific features; : Input data at time t, including building monitoring data; : The bias term of the i-th convolution kernel, used to adjust the output; : Rectified Linear Unit, used to activate the output of the convolutional layer and make it nonlinear;

[0022] Then, the time series is modeled through LSTM; combined with the time-dependent processing formula of the LSTM model: ;

[0023] : The hidden state at the current moment; : The hidden state of the previous moment, used to maintain the dependency relationship in the time series; : The features obtained from the convolutional layer at the current moment; : The weight matrix of LSTM controls the mapping between hidden state and input; : LSTM bias term; : Sigmoid activation function, used for nonlinear processing of hidden state updates.

[0024] Preferably, the adaptive time-varying Bayesian network module is used for dynamic risk assessment, and a time adjustment factor is introduced to adaptively adjust the conditional probability of different time periods. The formula is: ;

[0025] : The conditional probability of event A occurring when event B occurs at time t; : The conditional probability of event B occurring at time t given event A occurs; : The prior probability of event A at time t; : The marginal probability of event B at time t; : Time adjustment factor, which is dynamically adjusted with the monitoring data to adapt to the risk changes of building structures in different time periods.

[0026] Preferably, the random forest regression module adopts a random forest regression model, which is used to process the nonlinear characteristics of multi-source data of building structures, and obtain the final prediction by weighted average of the results of multiple decision trees. The formula is: ;

[0027] : The final prediction output represents the safety status of the building structure under the current monitoring conditions; : The number of decision trees, used to improve the generalization ability of the model; : The prediction result of the i-th decision tree is calculated based on the input data.

[0028] Preferably, the emergency response scheduling module adopts a genetic algorithm, evaluates the optimality of resource scheduling through a fitness function, and generates a new solution through crossover and mutation operations, specifically including the following steps:

[0029] Sp3.1, the fitness function is defined as: ; : Fitness function, which represents the optimization goal of the emergency resource scheduling plan; : The cost of the i-th type of emergency resources; : Decision variables of the i-th type of emergency resources; : The time or distance cost of the jth emergency path; : The selection variable of the jth emergency path;

[0030] Sp3.2, generate new individuals through crossover operation: ; : The newly generated individual represents the new emergency resource scheduling plan; : The first parent solution; : The second parent solution; : Crossover scale factor, used to determine the proportion of parent contribution, ranging from 0 to 1.

[0031] Preferably, the Sp1 collects monitoring data of building structures through multi-source sensors and also includes a global building health status monitoring module. The global building health status monitoring module adopts a dynamic graph convolutional network to dynamically capture the dependency of the building structure. The specific formula is: ;

[0032] The output feature matrix of the l+1th layer of graph convolution represents the updated features of the building structure nodes; : Degree matrix, indicating the number of connections for each node, is the normalized form of the degree matrix; : Adjacency matrix, which represents the connection relationship between each node in the building structure; : The input feature matrix of the lth layer, representing the node features of the building structure in the current state; : The weight matrix of the lth layer, used to learn feature representation; : Activation function, used for nonlinear processing.

[0033] Preferably, the Sp4 weighted fusion obtains the final building safety status prediction result, and the formula is: ;

[0034] : The final prediction result after fusion; : Prediction results of the improved LSTM model; : Prediction results of adaptive time-varying Bayesian network; : Prediction results of random forest regression model; : Model weights, which are dynamically adjusted based on the performance of each model, and .

[0035] Preferably, the method further comprises a multi-source data acquisition module, wherein the multi-source data acquisition module comprises:

[0036] Stress sensor unit, used to collect stress change data in building structure in real time;

[0037] Strain sensor unit, used to collect strain data of building materials under different load conditions;

[0038] Temperature and humidity sensor unit, used to monitor the ambient temperature and humidity around the building;

[0039] Vibration sensor units, used to monitor the response of building structures to earthquakes or external vibration conditions;

[0040] The data acquisition controller is used to synchronize and coordinate the data acquisition process of each sensor and transmit the data to the data preprocessing module.

[0041] Preferably, the global building health status monitoring module includes:

[0042] A dynamic graph construction unit is used to construct an adjacency matrix according to the connection relationship of each node of the building structure, and update the adjacency relationship according to the node status;

[0043] Feature extraction unit, which extracts the health status information of the building structure from the node features based on the graph convolutional network;

[0044] The node update unit updates each building node in real time based on the dynamic graph convolutional network, including the adaptive adjustment mechanism of the degree matrix and weight matrix;

[0045] The risk assessment unit combines the global building health status information with the dynamic Bayesian network to evaluate the overall risk level of the building structure and generate corresponding warnings.

[0046] Preferably, the Sp1 collects monitoring data of the building structure through a multi-source sensor, and the monitoring data is preprocessed. The data preprocessing steps are as follows:

[0047] Sp1.1, data denoising, using wavelet transform to filter noise from monitoring data, the formula of the wavelet transform is:

[0048] : Wavelet transform coefficient, indicating the change of signal at different scales; : original sensor signal; : mother wavelet function; : scale factor; : translation factor;

[0049] Sp1.2, data standardization, use Z-score standardization method to standardize the denoised data, the formula is:

[0050] : Standardized data; : The original data after denoising; : the mean of the data; : standard deviation of the data.

[0051] Beneficial effects:

[0052] The present invention provides a construction engineering safety monitoring and dynamic assessment method based on BIM technology. It has the following beneficial effects:

[0053] 1. The present invention monitors the health status of building structures in real time through monitoring data collection of multi-source sensors and multi-model fusion prediction framework, and effectively captures short-term and long-term dependencies in the data. The combined use of multiple models improves the accuracy of prediction, especially when dealing with complex time series data and nonlinear relationships, effectively reducing the errors that may be caused by a single model.

[0054] 2. The present invention introduces an adaptive time-varying Bayesian network, which dynamically adjusts the conditional probability through the time adjustment factor, so as to continuously and dynamically evaluate the risk of the building structure. When the building is in a potentially dangerous state, the system can conduct risk assessment and issue early warnings in a timely manner, helping managers take timely measures to prevent the occurrence of building disasters.

[0055] 3. The present invention uses genetic algorithms to dispatch emergency resources. This solution can generate the optimal emergency resource dispatch plan and path planning when an emergency occurs in a building. Compared with traditional static emergency dispatch, this solution dynamically optimizes the speed of emergency response and resource utilization, ensuring that resources can quickly reach the risk area and effectively intervene.

[0056] 4. The present invention can handle different types of monitoring data (such as stress, strain, temperature and humidity, vibration, etc.), and can also adaptively adjust the evaluation criteria and prediction mode through multi-model fusion. This means that no matter what environment and state the building is in, the system can adjust its prediction strategy according to real-time data to adapt to the complex and changing building environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a system diagram of the present invention;

[0058] Figure 2 It is the LSTM prediction graph of the present invention;

[0059] Figure 3 is the random forest prediction graph of the present invention;

[0060] Figure 4 is a multi-model prediction graph of the present invention;

[0061] Figure 5 It is a data iteration optimization graph of the present invention;

[0062] Figure 6 It is a data processing flow chart of the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Specific embodiment one:

[0065] As shown in the figure, the construction engineering safety monitoring and dynamic assessment method based on BIM technology includes the following steps:

[0066] Sp1, collect monitoring data of building structures through multi-source sensors, including environmental parameters and building parameters;

[0067] Sp2. Use a multi-model fusion prediction framework to dynamically predict and evaluate the structural health status of buildings. The multi-model fusion prediction framework includes the following parts:

[0068] Sp2.1, improved multi-scale long short-term memory network module to capture short-term and long-term dependency information in building structure monitoring data;

[0069] Sp2.2, Adaptive Time-Varying Bayesian Network Module, used to dynamically adjust the conditional probability in risk assessment according to changes in environmental conditions and building loads;

[0070] Sp2.3, random forest regression module, is used to process nonlinear high-dimensional data in building monitoring and perform multi-dimensional feature fusion;

[0071] Sp3, emergency response dispatch module, used to handle emergency situations;

[0072] Sp4, based on the data output of Sp1-Sp3, obtains the final building safety status prediction result through weighted fusion.

[0073] The improved multi-scale long short-term memory network module adopts an improved multi-scale long short-term memory network model. The improved multi-scale long short-term memory network model is as follows:

[0074] : Output features of the i-th convolution kernel; : The weight matrix of the i-th convolution kernel, representing the parameters used to extract specific features; : Input data at time t, including building monitoring data; : The bias term of the i-th convolution kernel, used to adjust the output; : Rectified Linear Unit, used to activate the output of the convolutional layer and make it nonlinear;

[0075] Then, the time series is modeled through LSTM; combined with the time-dependent processing formula of the LSTM model:

[0076] : The hidden state at the current moment; : The hidden state of the previous moment, used to maintain the dependency relationship in the time series; : The features obtained from the convolutional layer at the current moment; : The weight matrix of LSTM controls the mapping between hidden state and input; : LSTM bias term; : Sigmoid activation function, used for nonlinear processing of hidden state updates.

[0077] The adaptive time-varying Bayesian network module is used for dynamic risk assessment. The time adjustment factor is introduced to adaptively adjust the conditional probability of different time periods. The formula is:

[0078] : The conditional probability of event A occurring when event B occurs at time t; : The conditional probability of event B occurring at time t given event A occurs; : The prior probability of event A at time t; : The marginal probability of event B at time t; : Time adjustment factor, which is dynamically adjusted with the monitoring data to adapt to the risk changes of building structures in different time periods.

[0079] The random forest regression module uses a random forest regression model, which is used to process the nonlinear characteristics of multi-source data of building structures and obtain the final prediction through the weighted average of the results of multiple decision trees. The formula is: ;

[0080] : The final prediction output represents the safety status of the building structure under the current monitoring conditions; : The number of decision trees, used to improve the generalization ability of the model; : The prediction result of the i-th decision tree is calculated based on the input data.

[0081] The emergency response scheduling module uses a genetic algorithm to evaluate the optimality of resource scheduling through a fitness function and generates new solutions through crossover and mutation operations. Specifically, it includes the following steps:

[0082] Sp3.1, the fitness function is defined as:

[0083] : Fitness function, which represents the optimization goal of the emergency resource scheduling plan; : The cost of the i-th type of emergency resources; : Decision variables of the i-th type of emergency resources; : The time or distance cost of the jth emergency path; : The selection variable of the jth emergency path;

[0084] Sp3.2, generate new individuals through crossover operation: ;

[0085] : The newly generated individual represents the new emergency resource scheduling plan; : The first parent solution; : The second parent solution; : Crossover scale factor, used to determine the proportion of parent contribution, ranging from 0 to 1.

[0086] Sp1 collects monitoring data of building structures through multi-source sensors and also includes a global building health status monitoring module. The global building health status monitoring module uses a dynamic graph convolutional network to dynamically capture the dependencies of building structures. The specific formula is:

[0087] The output feature matrix of the l+1th layer of graph convolution represents the updated features of the building structure nodes; : Degree matrix, indicating the number of connections for each node, is the normalized form of the degree matrix; : Adjacency matrix, which represents the connection relationship between each node in the building structure; : The input feature matrix of the lth layer, representing the node features of the building structure in the current state; : The weight matrix of the lth layer, used to learn feature representation; : Activation function, used for nonlinear processing.

[0088] The final building safety status prediction result is obtained by Sp4 weighted fusion, and the formula is:

[0089] : The final prediction result after fusion; : Prediction results of the improved LSTM model; : Prediction results of adaptive time-varying Bayesian network; : Prediction results of random forest regression model; : Model weights, which are dynamically adjusted based on the performance of each model, and .

[0090] The method further includes a multi-source data acquisition module, the multi-source data acquisition module including:

[0091] Stress sensor unit, used to collect stress change data in building structure in real time;

[0092] Strain sensor unit, used to collect strain data of building materials under different load conditions;

[0093] Temperature and humidity sensor unit, used to monitor the ambient temperature and humidity around the building;

[0094] Vibration sensor units, used to monitor the response of building structures to earthquakes or external vibration conditions;

[0095] The data acquisition controller is used to synchronize and coordinate the data acquisition process of each sensor and transmit the data to the data preprocessing module.

[0096] The global building health monitoring module includes:

[0097] A dynamic graph construction unit is used to construct an adjacency matrix according to the connection relationship of each node of the building structure, and update the adjacency relationship according to the node status;

[0098] Feature extraction unit, which extracts the health status information of the building structure from the node features based on the graph convolutional network;

[0099] The node update unit updates each building node in real time based on the dynamic graph convolutional network, including the adaptive adjustment mechanism of the degree matrix and weight matrix;

[0100] The risk assessment unit combines the global building health status information with the dynamic Bayesian network to evaluate the overall risk level of the building structure and generate corresponding warnings.

[0101] Sp1 collects monitoring data of building structures through multi-source sensors and performs data preprocessing on the monitoring data. The data preprocessing steps are as follows:

[0102] Sp1.1, data denoising, using wavelet transform to filter noise from monitoring data, the formula of wavelet transform is:

[0103] : Wavelet transform coefficient, indicating the change of signal at different scales; : original sensor signal; : mother wavelet function; : scale factor; : translation factor;

[0104] Sp1.2, data standardization, use Z-score standardization method to standardize the denoised data, the formula is:

[0105] : Standardized data; : The original data after denoising; : the mean of the data; : standard deviation of the data.

[0106] First, the multi-source data acquisition module is responsible for acquiring the monitoring data of the building structure in real time through the sensor network. These data cover the environmental parameters and structural parameters of the building, among which the environmental parameters include temperature, humidity, vibration, etc., while the structural parameters involve stress, strain, etc. These monitoring data are collected by various sensors, such as stress sensors, strain sensors, temperature and humidity sensors, vibration sensors, etc. In order to ensure the synchronization and consistency of the data, the data acquisition controller coordinates the work of multiple sensors and transmits the collected data to the data preprocessing module. In this module, the data processing first involves the denoising step, using wavelet transform to filter the original monitoring data to remove environmental noise and high-frequency interference. The wavelet transform uses the mother wavelet function to decompose the signals at different scales and generate wavelet transform coefficients, thereby extracting the detailed features of the signal and filtering out the noise. Subsequently, the denoised data is processed by the Z-score standardization method. This process enables data from different sources to be compared and analyzed at the same scale to ensure data consistency and availability.

[0107] After data preprocessing, the data is input into the multi-model fusion prediction framework. The framework consists of three main models: an improved multi-scale long short-term memory network (LSTM) model, an adaptive time-varying Bayesian network, and a random forest regression model. First, the improved multi-scale long short-term memory network (LSTM) model focuses on processing time series data, and is particularly suitable for capturing short-term and long-term dependencies in building structure monitoring data. The model extracts multi-scale features through a convolutional layer and inputs these features into the LSTM network for modeling. The main function of the convolutional layer is to extract local features in the monitoring data using multiple convolution kernels, perform feature mapping on the input data through convolution operations, and generate convolution feature maps. Then, the LSTM network performs time series modeling based on these feature maps. The core of the LSTM network is to be able to memorize information over a longer time span and control the flow of information through forget gates, input gates, and output gates, thereby effectively capturing the long-term and short-term dependencies in the building structure state.

[0108] Secondly, the adaptive time-varying Bayesian network is used to dynamically adjust the conditional probability in the risk assessment process. The Bayesian network is a tool for probabilistic reasoning that can adjust the conditional probability according to changes in environmental conditions and building loads. In building structure monitoring, as the environment and load change, the health of the building structure will also fluctuate. Therefore, the adaptive time-varying Bayesian network dynamically adjusts the conditional probability by introducing a time adjustment factor, so that the assessment results can reflect the changes in the building status in real time. Specifically, at each moment, the system will update the conditional probability of the Bayesian network based on the real-time monitoring data. The conditional probability reflects the possibility of other events occurring under the premise of a certain building event. The time adjustment factor plays the role of adaptive adjustment. It adjusts according to the dynamic changes of the monitoring data to ensure that the Bayesian network model accurately assesses the risk in different time periods.

[0109] The random forest regression model is mainly used to process nonlinear high-dimensional data in building structure monitoring. The model is based on an integrated algorithm of multiple decision trees, and generates the final prediction by weighted averaging the results of multiple decision trees. Each decision tree predicts the state of the building structure based on the input features, and the random forest regression model that integrates multiple decision trees can effectively handle complex nonlinear features and reduce the risk of overfitting of a single model. Finally, the health status prediction of the building structure is obtained by weighted averaging the outputs of multiple decision trees.

[0110] While predicting the health status of building structures, the system integrates an emergency response scheduling module to deal with resource scheduling in building emergencies. This module uses genetic algorithms to optimize emergency resource allocation. Genetic algorithms are a search algorithm that simulates the biological evolution process and evaluates the pros and cons of resource allocation schemes through fitness functions. The fitness function comprehensively considers factors such as emergency resource costs, path time or distance to ensure that resources can be quickly and efficiently scheduled in emergency situations. The genetic algorithm generates new resource scheduling schemes through crossover and mutation operations. The crossover operation combines the characteristics of the two parent schemes to generate a new scheme, and the mutation operation increases the diversity of the solution space by randomly changing individual variables. After multiple generations of evolution and screening, the system finally generates the most adaptive emergency resource scheduling scheme.

[0111] In addition, the system also integrates a global building health status monitoring module, which uses a dynamic graph convolutional network (DGCN) to dynamically model the node dependencies of the building structure. Each node in the building structure represents a key part, such as a supporting column, beam, wall, etc. The connection relationship between the nodes constitutes the overall structure of the building. The dynamic graph convolutional network represents the connection between nodes through an adjacency matrix and extracts node features through graph convolution operations. The adjacency matrix describes the connection relationship between nodes, and the degree matrix represents the number of connections for each node. The graph convolutional network processes these matrices to extract the health status information of the building structure. The dynamic graph convolutional network can dynamically update the network weights and connection relationships according to node changes, ensuring that the system reflects the global health status of the building structure in real time. Finally, the system combines the global health status information with the adaptive time-varying Bayesian network to generate the overall risk assessment results of the building and issue an early warning signal.

[0112] The prediction results of all models are weighted and fused to obtain the final building safety status assessment results. The fusion process is based on the prediction results of LSTM, adaptive time-varying Bayesian network and random forest regression models, and the output of each model is integrated through weighted average. The weight of each model is dynamically adjusted according to its performance in historical data to ensure that the final prediction result combines the advantages of different models and improves the prediction accuracy and robustness. Specific embodiment 2:

[0114] As shown in the figure, the working plan of the construction project safety monitoring and dynamic assessment method based on BIM technology is as follows:

[0115] First, a virtual building structure model is constructed in the BIM modeling software. The model covers the main structural components of the building (such as beams, columns, walls, etc.) and integrates the building's information data (including materials, structural bearing capacity, design specifications, etc.). This model will serve as the basis for data collection and monitoring.

[0116] 1. Normal working status data: In the absence of external interference with the building, the building's stress, strain, temperature, humidity, and vibration data are collected through virtual sensors as baseline data for the building's normal state.

[0117] 2. Data affected by external interference: By setting external interference (such as earthquakes, strong winds, sudden temperature changes and other environmental changes), the response of the building structure under different external forces is simulated, and abnormal data of stress and strain are collected.

[0118] Specific sensors include:

[0119] Stress sensors: record stress changes at different locations on beams and columns.

[0120] Strain sensors: record the deformation of building materials under different load conditions.

[0121] Temperature and humidity sensor: records environmental conditions (temperature, humidity) outside the building.

[0122] Vibration sensors: record the effects of earthquakes or environmental vibrations on buildings.

[0123] The multi-source monitoring data obtained from the virtual sensor first passes through the data preprocessing module. The data preprocessing steps include:

[0124] De-noising: Wavelet transform is used to denoise the sensor data and filter out environmental noise and equipment interference.

[0125] Data standardization: The Zscore standardization method is used for sensor data from different sources so that all data have the same distribution characteristics, which is convenient for subsequent modeling processing.

[0126] Outlier detection: Detect outliers in monitoring data through statistical analysis and machine learning algorithms to ensure that abnormal data does not interfere with the model's prediction results.

[0127] The preprocessed data will be input into a multi-model fusion prediction framework for further analysis. The framework consists of three core models:

[0128] 1. Improved Multi-Scale LSTM Model: Specialized for analyzing time series data. This model learns features from monitoring data of buildings during normal operation and extracts multi-scale features through convolution operations to predict the future health of buildings. The weights of the convolutional layers are trained with monitoring data and combined with the long-term memory capacity of LSTM to capture dependencies between building structures in the long and short term.

[0129] 2. Adaptive Time-Varying Bayesian Network: This model dynamically adjusts conditional probabilities based on real-time monitoring data to assess changes in risk of building structures. The Bayesian network can estimate the likelihood of damage to buildings under different environmental conditions in real time through conditional probability reasoning. The time adjustment factor will be adaptively adjusted according to building loads and environmental changes.

[0130] 3. Random forest regression model: used to process nonlinear high-dimensional data. Random forest effectively captures the complex nonlinear relationships in building data by integrating multiple decision trees, and performs weighted average of the results of multiple decision trees to generate predictions on the health status of the building.

[0131] The three models work together to generate a prediction of the health status of the building structure. The prediction results of the models are integrated through weighted fusion, and the weighting coefficients are dynamically adjusted based on the historical performance of each model.

[0132] Once the model predicts that there is a potential risk in the building structure, the system will activate the emergency response scheduling module. This module uses genetic algorithms to generate the optimal resource scheduling plan. The genetic algorithm evaluates different emergency resource configuration plans through fitness functions and generates new emergency resource scheduling plans through crossover and mutation operations. Emergency resource scheduling includes rescue equipment, maintenance personnel, and emergency path planning.

[0133] The fitness function in the genetic algorithm will take into account factors such as the cost of emergency resources, response time, and path cost to ensure that emergency resources can quickly reach the damaged building area. By dynamically planning the emergency path through the A* algorithm, the system can choose the fastest emergency route and avoid obstacles or traffic jams around the building.

[0134] At the same time, the simulation system will use the dynamic graph convolutional network (DGCN) to monitor the global health status of the building. By representing the node connection relationship of the building structure through the adjacency matrix and degree matrix, DGCN can dynamically update the health status of the building and adjust the network weight in real time when the node changes. The global building health status information is combined with the Bayesian network to generate an overall risk assessment of the building.

[0135] After the experiment, the system will generate a safety status prediction report for the building structure, including monitoring data under normal conditions and when subject to external interference, risk assessment results, and emergency response scheduling plans. By comparing the simulation results with the expected values ​​of the actual state of the building, the accuracy and robustness of the multi-model fusion prediction framework can be verified.

[0136] Prediction accuracy: The prediction accuracy of the multi-model fusion prediction framework for building health status.

[0137] Emergency response efficiency: the response time and scheduling efficiency of the emergency resource scheduling solution generated by the genetic algorithm.

[0138] Global risk assessment: Accuracy of risk assessment of the overall building structure using dynamic graph convolutional networks.

[0139] Model fusion effect: By comparing the results of single model and multi-model fusion, the advantages of the fusion model can be evaluated.

[0140] Through this simulation experiment, the effectiveness of the construction project safety monitoring and dynamic assessment method based on BIM technology in practical applications can be verified. The working principles of each model have been specifically applied, and the collaborative work of data processing, model prediction, emergency response and global monitoring will provide a comprehensive assessment of the building safety status. This simulation solution shows how to use multi-source data collection and multi-model fusion technology to improve the safety monitoring and dynamic risk assessment capabilities of construction projects.

[0141] Data generation and preprocessing: Input building monitoring scene data, including stress, strain, temperature and humidity, and vibration. The data was denoised by sliding average and normalized by Z-score.

[0142] LSTM model: Based on the LSTM network training of building structure time series data, short-term and long-term dependency information is predicted.

[0143] Bayesian network dynamic update: Use a simple Bayesian network model to simulate the dynamic adjustment of conditional probability, mainly used for risk assessment.

[0144] Random Forest Regression: Use random forests for regression prediction of building health status and handle nonlinear data characteristics.

[0145] Model weighted fusion: The prediction results of LSTM, Bayesian network and random forest are weighted and integrated to obtain the final building health status prediction.

[0146] Result visualization and error evaluation: The prediction results of each model are displayed graphically, and the mean square error (MSE) is calculated to evaluate the model performance. Specific embodiment three:

[0148] Based on the technical solutions of the first and second specific embodiments, the following actual situations are further supplemented:

[0149] The deformation of pipelines and paths may lead to pipeline damage. Under complex environmental conditions, combined with the uncertainty of external forces, deformation, soil properties and other accidental change factors, ensuring the normal laying of pipelines and controlling deformation within an acceptable range is a necessary condition for the smooth progress of the project. For this project, the main purposes of monitoring include:

[0150] 1. By comparing the monitoring data with the predicted values, it is evaluated whether the construction process and parameters of the previous stage meet the expected requirements, and the subsequent construction process and progress are controlled accordingly to realize information construction;

[0151] 2. Timely and accurately forecast potential hazards that may threaten the safety of the foundation pit project itself and surrounding buildings, pipelines and other important environments, ensure the safety and normal use of the foundation pit structure and adjacent environment, and protect the interests of the owners and related society;

[0152] 3. Through monitoring, timely discover the leakage of the foundation pit water-stop curtain and notify the construction unit to make preparations for plugging the leakage to prevent large-scale sand gushing during the construction process;

[0153] 4. Feedback the on-site monitoring results to the design unit so that the design can be optimized according to the actual situation on site to achieve the goals of high quality, safety, economic rationality and fast construction.

[0154] Foundation pit monitoring frequency:

[0155] The following table is the frequency table of foundation pit monitoring:

[0156]

[0157] Note: 1. Observation points and reference points shall be set after the top of the slope is excavated and leveled, and shall be properly protected during the construction process; 2. The accuracy requirement for foundation pit deformation monitoring shall not be lower than the second-class accuracy.

[0158] In rainy weather or when there are situations that may lead to accelerated deformation (for example, the overload on the top of the slope increases significantly and exceeds the design allowable value), the observation frequency should be increased; once the foundation pit is backfilled, the observation can be stopped. If abnormal conditions such as a significantly accelerated deformation rate or cracks in the support structure are detected, the monitoring density should be increased immediately, and the monitoring data should be reported quickly to the supervision, design and construction personnel. If the deformation develops rapidly and signs of damage appear, continuous monitoring must be implemented to timely understand the deformation trend and accurately assess the safety status of the foundation pit.

[0159] The monitoring and control indicators of foundation pit support structure are shown in the table.

[0160] The following table shows the allowable values ​​and warning values ​​for monitoring of foundation pit support structures and surrounding environment:

[0161]

[0162] Note: 1. When the change rate of the monitoring project reaches the specified value in the table or exceeds 70% of the value for 3 consecutive days, or the cumulative value reaches 80% of the allowable value, an alarm should be issued in time. 2. The ground settlement around the foundation pit shall not affect the normal use of adjacent structures and various pipelines or the allowable value of differential settlement. 3. When the cumulative value of the overall inclination of the building reaches 2 / 1000 or the inclination speed is greater than 0.0001H / d for 3 consecutive days (H is the height of the building's load-bearing structure), an alarm should be issued.

[0163] If large deformation occurs during the monitoring process, additional monitoring points should be added to the surrounding areas to strengthen monitoring efforts.

[0164] The following table shows the deformation warning values ​​of various pipelines around the foundation pit:

[0165]

[0166] Deformation monitoring network layout: a certain number of benchmarks are laid out according to the specifications and design requirements to monitor the deformation of the working benchmarks, and are buried in stable and reliable locations away from the project point 3 times the excavation depth. Several leveling working benchmarks are placed in relatively stable places for various vertical displacement measurements in the surrounding area; several horizontal displacement working benchmarks are set at as far away as possible, and the horizontal displacement working benchmarks are regularly checked during the monitoring process. Through the proposed buried benchmarks, a monitoring network is formed for the surrounding environment of the foundation pit, and the monitoring points are effectively monitored. The monitoring benchmarks are re-measured in stages. After the benchmark network is re-measured, the stability of the benchmark should be analyzed and judged by combining and comparing the difference between the re-measured data and the last data. If the benchmark is found to have changed, it shall not be used as the starting point. When the benchmark is found to have changed after multiple re-measurements or a certain period of observation, the reference system should be re-selected and the original observation data should be used to re-calculate the previous results. The leveling benchmark uses a stainless steel wall leveling mark, which is set on a stable building; or a concrete ordinary leveling stone is used, which is set in a stable and reliable location. The working base point adopts displacement observation piers and has a forced centering device. The centering error shall not exceed 0.1mm. The level mark and displacement observation pier comply with the provisions of the "Building Deformation Measurement Code" JGJ8-2016. All benchmarks are selected as far as possible in the roadside, corners and other places that are easy to protect, and are marked with eye-catching signs. Communicate with each team to jointly protect the integrity of the benchmark points.

[0167] (1) Layout of plane monitoring and control network

[0168] A. For displacement observation of building foundation and site, two-level network is used: control network composed of control points, and extended network composed of observation points and control points measured in conjunction. The control network adopts angle measurement network, side measurement network, corner network or wire network, and the extended network and single-level network can be carried out in the form of intersection method (angle intersection, side intersection, corner intersection), baseline or attached wire.

[0169] B. For the first, second and third level control points when necessary, observation piers with forced centering devices are used. The centering error should not exceed 0.1mm. The control points should be easy to store, encrypt, expand and find for a long time, and the adjacent points should have good visibility and not be affected by side refraction.

[0170] C. Benchmark points, working base points, contact points, check points and orientation points should be arranged according to different network layout methods and configurations and the requirements of the "Code for Building Deformation Measurement". There should be no less than 3 benchmark points in each measuring area.

[0171] D. The benchmark point and working base point shall be buried in a place with a stable foundation, not easily damaged and far away from three times the depth of the foundation pit, and shall be covered with a protective cover.

[0172] (2) Layout of elevation control network

[0173] A. For survey areas with fewer buildings (structures), it is advisable to arrange the control points together with the observation points in a single layer. For large survey areas with many and scattered buildings, the network should be arranged in two layers, that is, the control network composed of control points, and the extended network composed of observation points and the control points connected to the network.

[0174] B. The control network is set as a closed loop, a node network or an attached elevation route, and the extended network is set as an attached (closed) elevation network.

[0175] C. There shall be no less than 3 leveling base points in each survey area.

[0176] D. The elevation reference point shall be buried in a stable and indestructible place far away from three times the depth of the foundation pit, and a bedrock mark shall be buried if necessary.

[0177] All benchmark points should be located in stable, reliable and easy-to-protect locations, marked clearly, and communicated with each team to jointly protect the integrity of the benchmark points.

[0178] Due to changes in the natural environment and human damage, it is inevitable that some points may change. In order to verify the stability of the benchmark points and ensure the reliability of each deformation measurement result, the benchmark points should be tested before each monitoring point observation. When the monitoring results suspect that the benchmark points may have changed, they should be re-tested immediately. The benchmark points should be re-tested regularly, and the re-testing time interval should be determined according to the stability of the points and changes in environmental conditions.

[0179] Monitoring of horizontal and vertical displacement of supporting structure: one monitoring point is located at 10-20m along each side of the foundation pit. The monitoring points for horizontal and vertical displacement are common points and do not need to be buried separately. Set at the top of the foundation pit pile or the top of the slope. The observation point must be located at a place where there is a clear line of sight and does not affect the construction, so it is easy to protect. The actual arrangement of the measuring points will be adjusted according to the on-site conditions. When burying the measuring point mark, attention should be paid to ensuring the clear line of sight between the measuring points. After the measuring point is buried, necessary protection should be carried out and obvious markings should be made.

[0180] At present, horizontal displacement monitoring and vertical displacement are mainly measured by high-precision measuring instruments (such as theodolites, levels, etc.) to measure deformation and settlement. In all engineering measurements, the most commonly used instruments are theodolites or full-drilling instruments. Theodolites are used to measure the coordinates (i.e., positions) of construction control points of terrain or structures and horizontal displacement during construction. Full-drilling instruments have the functions of both levels and theodolites. The forward intersection method, distance intersection method, and free station method are commonly used to monitor the two-dimensional (X, Y direction) horizontal displacement of foundation pits; the sight line method, small angle method, and distance measurement method are used to observe the horizontal unidirectional displacement of deformed bodies.

[0181] Surrounding roads (surface) and pipeline settlement monitoring: Road settlement points cannot be simply laid out on the road surface. Holes must be drilled in the road surface, casings must be buried, and steel measuring points must be placed in the casings. Fine sand must be poured around the pipes and between the casings and steel measuring points. The steel measuring rod is higher than the casing, and the protruding part is reserved according to the settlement estimate. An observation groove is set from the surface to the casing, and the groove depth is preset to 150mm. The top of the observation groove is closed with a metal cover to ensure accurate measurement points and normal traffic. When measuring, open the metal cover, put the indium steel ruler into the observation groove, and cover it after the measurement is completed.

[0182] It is planned to use the Leica LS15 electronic level to complete the closed leveling line or the attached leveling line in the BFFB or aBFFB measurement sequence, strictly in accordance with the national second-class leveling measurement method.

[0183] (1) Measurement method: Use a level and a matching leveling rod, and use the geometric leveling method to monitor vertical displacement according to the second-class leveling accuracy requirements. For all monitoring projects that use leveling (such as surrounding surface settlement, building settlement, etc.), the vertical displacement monitoring points should form a deformation monitoring network together with the benchmark points and working base points near the measurement area, and use a closed (attached) leveling route for observation. During the monitoring process, the monitoring benchmark points, monitoring instruments, monitoring personnel, monitoring routes and monitoring methods should be fixed.

[0184] (2) Assignment requirements:

[0185] ① In the early stage of monitoring, independently measure the reference point and deformation point three times according to the monitoring accuracy requirements, and take the average value as the initial value;

[0186] ② Before each monitoring, the benchmark points must be tested to verify their stability to ensure the reliability of the observation results;

[0187] ③ Observation should be carried out only after the level and ruler have passed the inspection, and avoid being carried out when the measuring point and ruler are vibrating;

[0188] ④ Try to choose to conduct observations at the same time every day, and adhere to the four fixed principles of observation, namely: fixed measurement personnel, fixed measurement point location, fixed measurement duration, and fixed measurement route;

[0189] According to the on-site conditions, flexible monitoring routes can be developed, and the measurement data can be uploaded to the monitoring platform after the measurement is completed.

[0190] According to relevant specifications, vertical displacement must meet the following accuracy requirements:

[0191] The following table shows the requirements for level accuracy and observation tolerance:

[0192]

[0193] The following table shows the vertical displacement monitoring accuracy requirements:

[0194]

[0195] According to the "Code for Building Deformation Measurement", the settlement observation points should be able to fully reflect the deformation characteristics of the building and foundation, and take into account the geological conditions and building structure characteristics. The points should be selected at the following locations:

[0196] 1) The four corners of the building, around the core tube, at large corners, and along the outer wall every 10m to 20m or every 2 to 3 columns.

[0197] 2) Both sides of the junction of high-rise and low-rise buildings, new and old buildings, and vertical and horizontal walls.

[0198] 3) Building cracks, both sides of post-cast joints and settlement joints, places with a large difference in foundation burial depth, the junction of artificial foundation and natural foundation, the boundary between different structures and the boundary between fill and excavation.

[0199] 4) For buildings with a width of 15m or less and in areas with complex geology or expansive soil, internal wall points should be set in the middle of the load-bearing internal partition walls, and ground points should be set in the center and around the indoor ground.

[0200] 5) On the foundation of each or part of a column of a frame structure building or along the horizontal line of the longitudinal axis.

[0201] 6) The four corners of raft foundation, box foundation and power equipment foundation, where the foundation form or buried depth changes, and both sides where the geological conditions change. According to the above requirements, the buildings around this station are frame structures, and the foundation is a raft foundation or a wall foundation. No major changes have been found in the geological conditions under the foundation. Therefore, the monitoring points are arranged around the buildings, and steel bars with a diameter of 10mm are anchored into the buildings or walls.

[0202] Deep horizontal displacement monitoring of supporting structures. According to the design and relevant specifications, deep horizontal displacement measurement points of pile bodies are arranged in the monitoring section, and inclinometer tubes are installed in the front row of double-row piles. The inclinometer tubes in the supporting pile bodies should be buried in time during the construction of supporting piles. The inclinometer tubes are fixed to the steel cage of the retaining wall by direct binding or setting clamps, and concrete is poured after the steel cage is put into the hole. The inclinometer tube is tied and buried with the steel cage of the supporting structure. The fixation of the inclinometer tube and the steel cage must be very stable to prevent the inclinometer tube and the steel cage from falling off when pouring concrete. At the same time, attention must be paid to the longitudinal torsion of the inclinometer tube. A very small torsion angle may cause the inclinometer probe to be stuck in the guide groove. At the same time, the installation and burial of the inclinometer tube of the retaining structure should comply with the following principles:

[0203] ① The bottom of the pipe should be level with or slightly lower than the bottom of the steel cage, and the top should reach the ground (or the top of the wall);

[0204] ② The inclinometer tube and the steel cage of the surrounding structure should be tied and buried, and the tying spacing should not be greater than 1.5m;

[0205] ③ The upper and lower pipes of the inclinometer tube should be well connected without any gaps, and the joints should be firmly fixed and sealed;

[0206] ④ When tying the pipe, the direction should be adjusted so that the pair of measuring grooves in the pipe are perpendicular to the measuring surface (i.e. parallel to the displacement direction);

[0207] ⑤ Seal the bottom and top, keep the inclinometer tube clean, unobstructed and straight, and tie a 300mm wide sack with wire at 0.75m from the top as a protective measure when the pile head is broken;

[0208] ⑥ Make clear markings and reliable protection measures during the monitoring period. If special circumstances such as missed burial or wrong burial occur, timely consultation with the supervision unit and the construction unit on supplementary burial should be carried out.

[0209] During the burial process, the inclinometer tube needs to be buried in accordance with the above requirements to ensure the direction of the slot and the casing protection of the inclinometer tube before pouring and before chiseling the pile head. If necessary, a certain amount of excess inclinometer tube needs to be pre-buried to ensure that valid data can be collected at the monitoring points.

[0210] The inclinometer is placed in the inclinometer tube, so that the guide wheel completely enters the guide groove. The direction should be that when the positive direction of the guide wheel is consistent with the positive direction of the measured displacement coordinate (+X), the measured value is positive, and the opposite is negative. When measuring with the inclinometer, first put the inclinometer into the bottom of the tube and measure from bottom to top. The inclinometer probe is equipped with a high-precision digital accelerometer, which has extremely high stability and completely eliminates the problem of pit vibration. There are two sets of guide wheels on the upper and lower parts of the probe, which are convenient for lifting and sliding along the guide groove of the inclinometer tube. The accelerometer of the probe can sense the inclination angle of the tube at each depth. Output a voltage signal, which is displayed on the panel of the reader. The voltage signal measured by the inclinometer is based on the inclinometer tube guide groove as the direction reference. At a certain depth, the sine function of the inclination angle on the standard spacing L of the upper and lower guide wheels of the inclinometer can be converted into horizontal displacement.

[0211] When the sensitive axis of the accelerometer is in the horizontal plane, the projection of vector g on the sensitive axis is zero, and the accelerometer output is zero. When there is an inclination angle between the sensitive axis of the accelerometer and the horizontal plane, the accelerometer outputs a voltage signal.

[0212] When these incremental horizontal deviations are accumulated and plotted as a curve starting from the bottom of the inclinometer hole, the result is a curve of the change in horizontal offset between the initial observation and any subsequent observations. This represents the deformation of the soil during this observation period, that is, the horizontal displacement. It is easy to see from this offset curve that an offset is occurring at a certain depth.

[0213] The deep horizontal displacement monitoring uses the INDEGEO-IN1000 borehole inclinometer. After the instrument collects data, the measurement data is imported and output via Bluetooth and sent to the automated monitoring platform.

[0214] The measuring holes are arranged according to the design requirements of the drawings, and the groundwater level monitoring wells are buried according to the following steps and requirements.

[0215] 1) Drilling: The water level observation hole is drilled with clean water. The diameter of the drill bit is Φ110 and the drilling is done in the vertical direction.

[0216] During the drilling process, relevant data such as stratum lithology and layer change depth, drilling time and initial water level should be recorded in a timely and accurate manner; drilling should be stopped after the borehole reaches the designed depth, the borehole should be cleaned in time, the borehole should be checked for patency, and cleaning records should be kept.

[0217] 2) Well pipe processing: The raw material of the well pipe is a PVC pipe with an inner diameter of Φ75 and a wall thickness of 2.5. In order to ensure the water permeability of the PVC pipe, a honeycomb Φ10 through hole is processed within the range of 0 to 4m under the PVC pipe. The circumferential spacing of the holes is 12mm, the axial spacing is 12mm, and the geotextile filter is wrapped.

[0218] 3) Well pipe placement: After the hole is drilled, the hole depth is verified to be correct and then the processed and inspected PVC well pipe with an inner diameter of Φ75 is lowered to ensure that the filter hole end faces downward. The water level observation hole should be 0.5m above the ground. A fixed measuring point mark should be set at the hole mouth and protected with a protective cover.

[0219] 4) Gravel filling: After the well pipe of the groundwater level observation hole is lowered into the hole, gravel with a particle size not greater than 5mm should be immediately filled around the outside of the well pipe.

[0220] 5) Well washing: After lowering the pipe and backfilling the gravel, wash the well with clean water in time.

[0221] 6) Check the water-stopping effect, seal the hole covers to prevent rainwater from entering, and install protective devices for the observation wells.

[0222] Before the system was connected to the platform, groundwater was monitored manually using a water level meter. When the probe descended in the well and touched the water surface, a red light came on accompanied by a buzzer sound. When the sensor touched the water, a red light flashed accompanied by an alarm sound, and the water level and depth were read out.

[0223] If the water level is monitored by an automated water level meter, it is necessary to install an automated water level meter at the pipe mouth after the water level pipe is buried, and the data is sent to the monitoring platform by wireless transmission. The automated water level meter is to slowly lower the water level gauge probe into the buried water pipe. When the probe touches the water surface, the reading of the measuring ruler at the top of the pipe is immediately read. The top position of the pipe corresponding to each reading should be consistent. Based on the elevation of the pipe top and the height difference between the pipe top and the ground, the elevation and burial depth of the groundwater level can be calculated.

[0224] The installation requirements of the anchor cable dynamometer are relatively strict, requiring that the anchor pad, upper and lower load-bearing plates, anchor cable dynamometer, anchor plate, tensioning jack, etc. should all be coaxial with the anchor cable hole. The anchor cable dynamometer is placed directly between the anchor pad and the anchor plate.

[0225] The VWA type vibrating wire anchor dynamometer is used. Several vibrating wire strain gauges are evenly distributed on the measuring steel cylinder. When the load causes the steel cylinder to deform, the strain gauge and the steel cylinder deform synchronously. The deformation causes the vibrating wire of the strain gauge to produce stress changes, thereby changing the vibration frequency of the vibrating wire. The electromagnetic coil excites the vibrating wire and measures its vibration frequency. The frequency signal is transmitted to the vibrating wire acquisition device via a cable, and the strain amount that causes the deformation of the measuring steel cylinder can be measured. Substituting the calibration coefficient can calculate the load value felt by the anchor dynamometer.

[0226] The internal force of the anchor cable is monitored by the anchor cable axial force meter. If automatic monitoring is implemented, a data acquisition terminal must be added. The cable of the anchor cable stress is connected to the automatic acquisition module, and the monitoring data is sent to the monitoring platform through wireless transmission.

[0227] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprising a reference structure" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0228] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A construction engineering safety monitoring and dynamic assessment method based on BIM technology, characterized by: The method comprises the following steps: Sp1, collect monitoring data of building structures through multi-source sensors, including environmental parameters and building parameters; Sp2. Use a multi-model fusion prediction framework to dynamically predict and evaluate the structural health status of the building. The multi-model fusion prediction framework includes the following parts: Sp2.1, improved multi-scale long short-term memory network module to capture short-term and long-term dependency information in building structure monitoring data; Sp2.2, Adaptive Time-Varying Bayesian Network Module, used to dynamically adjust the conditional probability in risk assessment according to changes in environmental conditions and building loads; Sp2.3, random forest regression module, is used to process nonlinear high-dimensional data in building monitoring and perform multi-dimensional feature fusion; Sp3, emergency response dispatch module, used to handle emergency situations; Sp4, based on the data output of Sp1-Sp3, obtains the final building safety status prediction result through weighted fusion.

2. The construction engineering safety monitoring and dynamic assessment method based on BIM technology according to claim 1 is characterized by: The improved multi-scale long short-term memory network module adopts an improved multi-scale long short-term memory network model, and the improved multi-scale long short-term memory network model is as follows: ； : Output features of the i-th convolution kernel; : The weight matrix of the i-th convolution kernel, representing the parameters used to extract specific features; : Input data at time t, including building monitoring data; : The bias term of the i-th convolution kernel, used to adjust the output; : Rectified linear unit, used to activate the output of the convolutional layer and make it nonlinear.

3. The construction engineering safety monitoring and dynamic assessment method based on BIM technology according to claim 1 is characterized by: The adaptive time-varying Bayesian network module is used for dynamic risk assessment. The time adjustment factor is introduced to adaptively adjust the conditional probability of different time periods. The formula is: ; : The conditional probability of event A occurring when event B occurs at time t; : The conditional probability of event B occurring at time t given event A occurs; : The prior probability of event A at time t; : The marginal probability of event B at time t; : Time adjustment factor, which is dynamically adjusted with the monitoring data to adapt to the risk changes of building structures in different time periods.

4. The construction engineering safety monitoring and dynamic assessment method based on BIM technology according to claim 1 is characterized by: The random forest regression module adopts a random forest regression model, which is used to process the nonlinear characteristics of multi-source data of building structures and obtain the final prediction through weighted average of the results of multiple decision trees.

5. The construction engineering safety monitoring and dynamic assessment method based on BIM technology according to claim 1 is characterized by: The emergency response scheduling module adopts a genetic algorithm to evaluate the optimality of resource scheduling through a fitness function and generates a new solution through crossover and mutation operations, specifically including the following steps: Sp3.1, the fitness function is defined as: ； : Fitness function, which represents the optimization goal of the emergency resource scheduling plan; : The cost of the i-th type of emergency resources; : Decision variables of the i-th type of emergency resources; : The time or distance cost of the jth emergency path; : The selection variable of the jth emergency path; Sp3.2, generate new individuals through crossover operation: ; : The newly generated individual represents the new emergency resource scheduling plan; : The first parent solution; : The second parent solution; : Crossover scale factor, used to determine the proportion of parent contribution, ranging from 0 to 1.

6. The construction engineering safety monitoring and dynamic assessment method based on BIM technology according to claim 1 is characterized by: The Sp1 collects monitoring data of building structures through multi-source sensors and also includes a global building health status monitoring module. The global building health status monitoring module adopts a dynamic graph convolutional network to dynamically capture the dependencies of building structures.

7. The construction engineering safety monitoring and dynamic assessment method based on BIM technology according to claim 1 is characterized by: The Sp4 weighted fusion obtains the final building safety status prediction result.

8. The construction engineering safety monitoring and dynamic assessment method based on BIM technology according to claim 6 is characterized by: The global building health status monitoring module includes: A dynamic graph construction unit is used to construct an adjacency matrix according to the connection relationship of each node of the building structure, and update the adjacency relationship according to the node status; Feature extraction unit, which extracts the health status information of the building structure from the node features based on the graph convolutional network; The node update unit updates each building node in real time based on the dynamic graph convolutional network, including the adaptive adjustment mechanism of the degree matrix and weight matrix; The risk assessment unit combines the global building health status information with the dynamic Bayesian network to evaluate the overall risk level of the building structure and generate corresponding warnings.

9. The construction engineering safety monitoring and dynamic assessment method based on BIM technology according to claim 1 is characterized by: Sp1, collecting monitoring data of building structures through multi-source sensors, and performing data preprocessing on the monitoring data. The data preprocessing steps are as follows: Sp1.1, data denoising, using wavelet transform to filter noise from monitoring data; Sp1.2, data standardization processing, use Z-score standardization method to standardize the denoised data.

Images