A method for predicting apparent damage trend of existing buildings

Abstract

The present invention discloses a method for predicting the apparent damage trend of existing buildings, which relates to the field of building detection technology and includes the following steps: S1, defining damage characteristics and classification; S2, selecting appropriate sensors; S3, establishing a multi-dimensional data acquisition system; S4, designing a 4D network model; S5, model training and tuning; S6, constructing a dynamic association mechanism; S7, result verification and reliability assessment; S8, constructing a visualization and interactive platform; S9, continuous learning and model update mechanism. This method can capture changes in the apparent damage of buildings in a timely manner through real-time data input and dynamic updating of model status. The attention mechanism is used to strengthen the focus on key damage characteristics, thereby improving the accuracy and real-time performance of the prediction. Through this innovative method, the efficiency and accuracy of building damage monitoring can be significantly improved, providing more reliable data support for building safety management, and ultimately extending the service life of the building and reducing maintenance costs.

Description

Technical Field

[0001] The present invention relates to the technical field of building detection, and in particular to a method for predicting apparent damage trends of existing buildings. Background Art

[0002] With the acceleration of urbanization, the safety and durability of existing buildings have become a major societal issue. Existing building damage monitoring technologies primarily rely on regular manual inspections or simple sensor data analysis, resulting in delayed responses to damage status and a lack of real-time and comprehensiveness. Furthermore, traditional methods often struggle to effectively integrate multi-dimensional data and cannot dynamically capture changes in building damage over time and space, hindering the accurate prediction of damage trends.

[0003] To solve the above problems, we proposed a method for predicting the apparent damage trend of existing buildings. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method for predicting the apparent damage trend of existing buildings to solve the problems raised in the above background technology.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solution: a method for predicting the apparent damage trend of an existing building, comprising the following steps:

[0006] S1. Define injury characteristics and classification;

[0007] Identify the damage characteristics of existing buildings that need to be monitored, and develop damage classification standards through literature research and historical damage records to effectively distinguish different types of damage during modeling;

[0008] S2. Select the appropriate sensor;

[0009] Based on the damage characteristics to be monitored, appropriate sensors are selected to collect relevant data on the damage characteristics. Sensors include image recognition equipment, laser scanners, and strain sensors to ensure that the selected sensors can accurately reflect the damage status of existing buildings in real time.

[0010] S3. Establish a multi-dimensional data collection system;

[0011] Combine sensor data, environmental factors, and historical damage records to form a monitoring database, establish a multi-dimensional data collection system, and comprehensively analyze multi-dimensional data to understand the influencing factors and development trends of damage;

[0012] S4, design 4D network model;

[0013] After data processing is complete, a 4D network model suitable for building damage prediction is designed. The 4D network model combines spatiotemporal characteristics to dynamically capture changes in damage status over time and space. The 4D network model also introduces an attention mechanism to enhance the model's ability to focus on key damage features and improve prediction accuracy.

[0014] S5, model training and tuning;

[0015] The designed 4D network model is trained using the processed data, cross-validation is used to evaluate the performance of the model, and model parameters such as learning rate and batch size are adjusted based on the evaluation results;

[0016] S6, construction of dynamic association mechanism;

[0017] Build a dynamic association mechanism for damage state information, network models, and monitoring solutions. This mechanism updates the model state based on real-time data input and adjusts the monitoring solution through a feedback loop to ensure that the model always adapts to the latest building conditions.

[0018] S7, result verification and reliability assessment;

[0019] After the model training is completed, the prediction results are evaluated using an independent validation set;

[0020] S8, build visualization and interaction platform;

[0021] Build a visualization and interactive platform, using data visualization technology to present prediction results, real-time monitoring data, and historical damage records in the form of charts, and provide user interaction functions so that users can customize monitoring plans according to actual needs;

[0022] S9. Continuous learning and model updating mechanism.

[0023] Further optimizing the technical solution, in step S1, the damage characteristics of the existing building include appearance cracks, deformation, corrosion, shedding, water penetration and deterioration;

[0024] in,

[0025] Appearance cracks include crack width, depth, and direction;

[0026] Deformation includes displacement, tilt, and depression;

[0027] Corrosion includes corrosion degree and distribution;

[0028] Shedding includes material flaking and fragment distribution;

[0029] Water penetration includes water marks and humidity changes;

[0030] Deterioration includes loss of material strength and signs of aging.

[0031] To further optimize this technical solution, in step S3, the multi-dimensional data collection system includes a database system and a data processing platform;

[0032] Database system: Choose a suitable database management system such as MySQL or PostgreSQL to support large-scale data storage and query;

[0033] Data processing platform, using data processing platforms such as Apache Spark for data cleaning, integration and analysis, providing flexible analysis capabilities and efficient data processing capabilities;

[0034] The multi-dimensional data collection system collects multi-dimensional data into the database system through the data collection interface to ensure real-time update and accessibility of the data;

[0035] Using data analysis tools, statistical analysis and correlation analysis are performed on multi-dimensional data to identify potential influencing factors of damage and generate a damage development trend model.

[0036] To further optimize this technical solution, in step S4, the 4D network model combines time T, space X, Y, and feature dimension F to predict damage trends. The 4D network model consists of the following parts:

[0037] Input layer: Receives multi-dimensional data, including sensor data, environmental factors, and historical damage records;

[0038] Convolutional layer: used to extract spatial features and identify damage patterns;

[0039] Circular layer: captures time series data and identifies damage trends over time;

[0040] Attention mechanism: Enhances the model's focus on key damage features and improves prediction accuracy;

[0041] Output layer: outputs the prediction results of future damage status.

[0042] To further optimize this technical solution, the 4D network model includes the following processes when being constructed:

[0043] Assume a 4D tensor ,in, For time, is the spatial coordinate X axis, is the spatial coordinate Y axis, is the feature dimension;

[0044] The formula of the 4D network model is as follows:

[0045] ;

[0046] in, is the output of the convolutional layer, is the convolution kernel, and its size is set according to actual needs. is the bias term;

[0047] Use recurrent layers to process time series data:

[0048] ;

[0049] in, is the output of the recurrent layer, indicating the change of damage characteristics over time.

[0050] To further optimize this technical solution, the attention mechanism introduced by the 4D network model is as follows:

[0051] ;

[0052] in, is the attention weight, is the learned weight matrix, is the feature dimension;

[0053] Finally, the prediction result of the output layer is expressed as:

[0054] ;

[0055] in, For damage status prediction.

[0056] To further optimize this technical solution, in step S5, cross-validation is used to evaluate the performance of the model, including the following specific process:

[0057] Data segmentation: Divide the preprocessed dataset into training set, validation set and test set;

[0058] Model training: Use the training set to train the 4D network model;

[0059] Cross-validation: Use k-fold cross-validation, such as k=5, to divide the training set into k subsets, and perform k training cycles, each time using one subset as the validation set and the other subsets as the training set;

[0060] Performance evaluation: Calculate the performance indicators of the model through the validation set, such as root mean square error and accuracy;

[0061] Parameter tuning: adjust model parameters based on evaluation results;

[0062] Final training: After multiple rounds of training and tuning, the entire training set is trained using the finalized hyperparameters.

[0063] To further optimize this technical solution, in the model training, the loss function in the training process is selected as mean square error or cross entropy loss, and the formula is as follows:

[0064] ;

[0065] in, For loss, is the sample size, is the true value, is the predicted value;

[0066] The parameter tuning includes:

[0067] Learning rate: Use learning rate scheduling strategies such as cosine annealing to dynamically adjust the learning rate;

[0068] Batch size: Try different batch sizes to optimize training speed and stability;

[0069] Regularization: Add L1 or L2 regularization terms to prevent the model from overfitting.

[0070] To further optimize this technical solution, in step S6, the dynamic association mechanism includes the following specific processes when being constructed:

[0071] Real-time data input: real-time monitoring data input, including sensor data, environmental factors and historical damage records;

[0072] State update: When new data is input, the model state is updated and the new damage state is calculated using the following formula: :

[0073] ;

[0074] in, is the damage status at the current moment, is the damage status at the previous moment, is the damage change calculated based on the new input data;

[0075] Feedback loop: Compare the model's prediction results with the actual monitoring data and calculate the prediction error. The calculation formula is as follows:

[0076] ;

[0077] in, is the prediction error, is the actual monitoring value, is the model prediction value;

[0078] Monitoring plan adjustment: Dynamically adjust the monitoring plan based on the prediction error. If the error exceeds the set threshold, increase the sensor collection frequency or change the monitoring point to obtain more accurate data;

[0079] Regular model retraining: Set up regular model retraining to update model weights using the latest monitoring data to maintain accuracy and effectiveness.

[0080] To further optimize the technical solution, in step S7, the prediction results are compared with the actual monitoring results to analyze the accuracy and recall rate of the model to ensure the reliability of the prediction results of the model;

[0081] Long-term tracking and monitoring will be carried out to comprehensively evaluate the prediction results of the model in different time periods.

[0082] Compared with the existing technology, the present invention provides a method for predicting the apparent damage trend of existing buildings, which has the following beneficial effects:

[0083] This method for predicting apparent damage trends in existing buildings uses real-time data input and dynamic updates of model states to promptly capture changes in apparent damage. It also leverages an attention mechanism to enhance focus on key damage characteristics, improving both the accuracy and real-time nature of predictions. This innovative approach significantly improves the efficiency and accuracy of building damage monitoring, providing more reliable data support for building safety management, ultimately extending the lifespan of buildings and reducing maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 This is a flow chart of a method for predicting apparent damage trends of existing buildings proposed by the present invention;

[0085] Figure 2 This is a schematic diagram of a multi-dimensional data acquisition system in a method for predicting apparent damage trends of existing buildings proposed by the present invention;

[0086] Figure 3 This is a schematic diagram of the construction of a 4D network model in a method for predicting apparent damage trends of existing buildings proposed in the present invention;

[0087] Figure 4 This is a flow chart of model training and optimization in a method for predicting apparent damage trends of existing buildings proposed by the present invention;

[0088] Figure 5 This is a schematic diagram of the construction of a dynamic correlation mechanism in a method for predicting the apparent damage trend of existing buildings proposed in the present invention. DETAILED DESCRIPTION

[0089] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described 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 making creative efforts are within the scope of protection of the present invention.

[0090] Example:

[0091] See also Figure 1 A method for predicting the apparent damage trend of an existing building comprises the following steps:

[0092] S1. Define injury characteristics and classification

[0093] Identify the damage characteristics of existing buildings that need to be monitored, and develop damage classification standards through literature research and historical damage records to effectively distinguish different types of damage when modeling.

[0094] In this embodiment, in step S1, the damage characteristics of the existing building include appearance cracks, deformation, corrosion, shedding, water penetration and deterioration.

[0095] in,

[0096] Appearance cracks include crack width, depth, and direction;

[0097] Deformation includes displacement, tilt, and depression;

[0098] Corrosion includes corrosion degree and distribution;

[0099] Shedding includes material flaking and fragment distribution;

[0100] Water penetration includes water marks and humidity changes;

[0101] Deterioration includes loss of material strength and signs of aging.

[0102] Based on the above damage characteristics, appropriate monitoring methods are selected. The corresponding monitoring methods for damage characteristics are shown in Table 1. Table 1 can help clearly define damage characteristics and monitoring methods, facilitating subsequent data analysis and model construction.

[0103]

[0104] Table 1

[0105] S2. Choose the right sensor

[0106] Based on the damage characteristics that need to be monitored, appropriate sensors are selected to collect relevant data on the damage characteristics. The sensors include image recognition equipment, laser scanners, and strain sensors to ensure that the selected sensors can reflect the damage status of existing buildings in real time and accurately.

[0107] S3. Establish a multi-dimensional data collection system

[0108] Combining sensor data, environmental factors and historical damage records, a monitoring database is formed, a multi-dimensional data collection system is established, and multi-dimensional data is comprehensively analyzed to understand the influencing factors and development trends of damage.

[0109] like Figure 2 As shown, in this embodiment, in step S3, the multi-dimensional data collection system includes a database system and a data processing platform;

[0110] Database system: Choose a suitable database management system such as MySQL or PostgreSQL to support large-scale data storage and query;

[0111] Data processing platform, using data processing platforms such as Apache Spark for data cleaning, integration and analysis, providing flexible analysis capabilities and efficient data processing capabilities;

[0112] The multi-dimensional data collection system collects multi-dimensional data into the database system through the data collection interface to ensure real-time update and accessibility of the data;

[0113] Using data analysis tools, statistical analysis and correlation analysis are performed on multi-dimensional data to identify potential influencing factors of damage and generate a damage development trend model.

[0114] In this embodiment, a multi-dimensional data collection system is shown in Table 2.

[0115] Table 2

[0116] S4. Design 4D network model

[0117] After the data processing is ready, a 4D network model suitable for building damage prediction is designed. The 4D network model combines spatiotemporal characteristics to dynamically capture the changes in damage status over time and space. The 4D network model also introduces an attention mechanism to enhance the model's ability to focus on key damage features and improve prediction accuracy.

[0118] like Figure 3 As shown, in this embodiment, in step S4, the 4D network model combines time T, space X, Y, and feature dimension F to perform damage trend prediction. The 4D network model consists of the following parts:

[0119] Input layer: Receives multi-dimensional data, including sensor data, environmental factors, and historical damage records;

[0120] Convolutional layer: used to extract spatial features and identify damage patterns;

[0121] Circular layer: captures time series data and identifies damage trends over time;

[0122] Attention mechanism: Enhances the model's focus on key damage features and improves prediction accuracy;

[0123] Output layer: outputs the prediction results of future damage status.

[0124] The construction of the 4D network model includes the following processes:

[0125] Assume a 4D tensor ,in, For time, is the spatial coordinate X axis, is the spatial coordinate Y axis, is the feature dimension;

[0126] The formula of the 4D network model is as follows:

[0127] ;

[0128] in, is the output of the convolutional layer, is the convolution kernel, and its size is set according to actual needs. is the bias term;

[0129] Use recurrent layers to process time series data:

[0130] ;

[0131] in, is the output of the recurrent layer, indicating the change of damage characteristics over time.

[0132] The attention mechanism introduced by the 4D network model is as follows:

[0133] ;

[0134] in, is the attention weight, is the learned weight matrix, is the feature dimension;

[0135] Finally, the prediction result of the output layer is expressed as:

[0136] ;

[0137] in, For damage status prediction.

[0138] Based on the above model, when using:

[0139] Step 1: Data Preparation

[0140] During data preparation, ensure that the data format conforms to 4D tensor The data should be normalized for better calculation in the model.

[0141] Step 2: Model training

[0142] Use the defined model structure and train it with historical data. Use the back propagation algorithm to optimize the model parameters. 、 、 . Cross-validation can be used to evaluate the model effect during training.

[0143] Step 3: Dynamic Input

[0144] After the model training is completed, the real-time monitoring data can be dynamically input into the model. Update the model state. Each time new data is input, new features are extracted through the convolutional layer, the temporal state is updated through the recurrent layer, and the key features are strengthened using the attention mechanism.

[0145] Step 4: Predict output

[0146] Finally, the prediction results are obtained through the output layer , providing future damage trend predictions. This process can provide an important basis for subsequent maintenance decisions and risk assessments.

[0147] S5. Model training and tuning

[0148] The designed 4D network model is trained using the processed data, and cross-validation is used to evaluate the performance of the model. The model parameters, such as learning rate and batch size, are adjusted based on the evaluation results.

[0149] like Figure 4 As shown, in this embodiment, in step S5, cross-validation is used to evaluate the performance of the model, including the following specific processes:

[0150] Data splitting: The preprocessed dataset is divided into training set, validation set and test set. Usually 70% is used for training, 15% for validation and 15% for testing to ensure the generalization ability of the model.

[0151] Model training: Use the training set to train the 4D network model;

[0152] Cross-validation: Use k-fold cross-validation, for example, k=5, to divide the training set into k subsets. Repeat the training cycle k times, using one subset as the validation set each time and the other subsets as the training set to more comprehensively evaluate the model performance and avoid overfitting.

[0153] Performance evaluation: Calculate the model's performance indicators, such as root mean square error and accuracy, using the validation set to ensure the model's predictive ability. At the same time, plot a learning curve to observe changes in training and validation errors to determine whether further tuning is needed.

[0154] Parameter tuning: adjust model parameters based on evaluation results;

[0155] Final training: After multiple rounds of training and tuning, the entire training set is trained using the finalized hyperparameters.

[0156] Through the above steps, the predictive ability of the 4D network model can be effectively improved, providing reliable support for the monitoring and assessment of building damage trends.

[0157] In the model training, the loss function in the training process selects mean square error or cross entropy loss, and the formula is as follows:

[0158] ;

[0159] in, For loss, is the sample size, is the true value, is the predicted value;

[0160] The parameter tuning includes:

[0161] Learning rate: Use learning rate scheduling strategies such as cosine annealing to dynamically adjust the learning rate;

[0162] Batch size: Try different batch sizes to optimize training speed and stability;

[0163] Regularization: Add L1 or L2 regularization terms to prevent the model from overfitting.

[0164] S6. Construction of dynamic association mechanism

[0165] A dynamic association mechanism is constructed to dynamically associate damage status information, network models, and monitoring plans. The mechanism updates the model status based on real-time data input and adjusts the monitoring plan through a feedback loop to ensure that the model always adapts to the latest building conditions.

[0166] like Figure 5As shown, in this embodiment, in step S6, the dynamic association mechanism includes the following specific processes when being constructed:

[0167] Real-time data input: real-time monitoring data input, including sensor data, environmental factors and historical damage records;

[0168] State update: When new data is input, the model state is updated and the new damage state is calculated using the following formula: :

[0169] ;

[0170] in, is the damage status at the current moment, is the damage status at the previous moment, is the damage change calculated based on the new input data;

[0171] Feedback loop: Compare the model's prediction results with the actual monitoring data and calculate the prediction error. The calculation formula is as follows:

[0172] ;

[0173] in, is the prediction error, is the actual monitoring value, is the model prediction value; this error is used to adjust the monitoring plan and model parameters to ensure their continuous optimization;

[0174] Monitoring plan adjustment: Dynamically adjust the monitoring plan based on the predicted error. If the error exceeds the set threshold, increase the sensor acquisition frequency or change the monitoring point to obtain more accurate data. This adjustment ensures that the monitoring plan adapts to changes in building conditions and detects potential damage in a timely manner.

[0175] Regular model retraining: Set up regular model retraining and use the latest monitoring data to update the model weights to maintain its accuracy and effectiveness. This retraining can be based on the performance evaluation results of the model to effectively improve the prediction ability.

[0176] Through a dynamic correlation mechanism, damage status information, network models, and monitoring solutions can be interconnected in real time, ensuring the accuracy and timeliness of building damage prediction. The implementation of this mechanism will provide strong support for building maintenance decision-making.

[0177] S7. Result Verification and Reliability Evaluation

[0178] After the model training is completed, the prediction results are evaluated using an independent validation set.

[0179] In this embodiment, in step S7, the prediction results are compared with the actual monitoring results to analyze the accuracy and recall rate of the model to ensure the reliability of the prediction results of the model;

[0180] Long-term tracking and monitoring will be carried out to comprehensively evaluate the prediction results of the model in different time periods.

[0181] S8. Build a visualization and interactive platform

[0182] Build a visualization and interactive platform, use data visualization technology to present prediction results, real-time monitoring data and historical damage records in the form of charts, and provide user interaction functions so that users can customize monitoring plans according to actual needs.

[0183] S9. Continuous learning and model updating mechanism

[0184] Establish a continuous learning and model update mechanism to regularly retrain and optimize the model by continuously introducing new monitoring data and feedback.

[0185] In this embodiment, this mechanism ensures that the model remains efficient and accurate over time, adapting to changing building conditions and external environments. This self-updating capability is key to achieving effective long-term predictions.

[0186] The beneficial effects of the present invention are:

[0187] This method for predicting apparent damage trends in existing buildings uses real-time data input and dynamic updates of model states to promptly capture changes in apparent damage. It also leverages an attention mechanism to enhance focus on key damage characteristics, improving both the accuracy and real-time nature of predictions. This innovative approach significantly improves the efficiency and accuracy of building damage monitoring, providing more reliable data support for building safety management, ultimately extending the lifespan of buildings and reducing maintenance costs.

[0188] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0189] While 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 these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the apparent damage trend of existing buildings, characterized in that: The following steps are involved: S1. Define injury characteristics and classification; Identify the damage characteristics of existing buildings that need to be monitored, and develop damage classification standards through literature research and historical damage records to effectively distinguish different types of damage during modeling; S2. Select appropriate sensor; Based on the damage characteristics to be monitored, appropriate sensors are selected to collect relevant data on the damage characteristics. Sensors include image recognition equipment, laser scanners, and strain sensors to ensure that the selected sensors can accurately reflect the damage status of existing buildings in real time. S3. Establish a multi-dimensional data collection system; Combine sensor data, environmental factors, and historical damage records to form a monitoring database, establish a multi-dimensional data collection system, and comprehensively analyze multi-dimensional data to understand the influencing factors and development trends of damage; S4, design 4D network model; After data processing is complete, a 4D network model suitable for building damage prediction is designed. The 4D network model combines spatiotemporal characteristics to dynamically capture changes in damage status over time and space. The 4D network model also introduces an attention mechanism to enhance the model's ability to focus on key damage features and improve prediction accuracy. S5, model training and tuning; The designed 4D network model is trained using the processed data, and the performance of the model is evaluated using cross-validation. The model parameters, including learning rate and batch size, are adjusted based on the evaluation results. S6, construction of dynamic association mechanism; Build a dynamic association mechanism for damage state information, network models, and monitoring solutions. This mechanism updates the model state based on real-time data input and adjusts the monitoring solution through a feedback loop to ensure that the model always adapts to the latest building conditions. S7, result verification and reliability assessment; After the model training is completed, the prediction results are evaluated using an independent validation set; S8, build visualization and interaction platform; Build a visualization and interactive platform, using data visualization technology to present prediction results, real-time monitoring data, and historical damage records in the form of charts, and provide user interaction functions so that users can customize monitoring plans according to actual needs; S9, continuous learning and model updating mechanism; Establish a continuous learning and model update mechanism to regularly retrain and optimize the model by continuously introducing new monitoring data and feedback.

2. A method for predicting apparent damage trend of existing buildings according to claim 1, characterized in that: In step S1, the damage characteristics of the existing building include appearance cracks, deformation, corrosion, shedding, water penetration and deterioration; in, Appearance cracks include crack width, depth, and direction; Deformation includes displacement, tilt, and depression; Corrosion includes corrosion degree and distribution; Shedding includes material flaking and fragment distribution; Water penetration includes water marks and humidity changes; Deterioration includes loss of material strength and signs of aging.

3. The method for predicting apparent damage trend of existing buildings according to claim 1, characterized in that: In step S3, the multi-dimensional data collection system includes a database system and a data processing platform; Database system: Choose a suitable database management system, including MySQL and PostgreSQL, to support large-scale data storage and query; Data processing platform, including Apache Spark, is used for data cleaning, integration, and analysis, providing flexible analytical capabilities and efficient data processing capabilities; The multi-dimensional data collection system collects multi-dimensional data into the database system through the data collection interface to ensure real-time update and accessibility of the data; Using data analysis tools, statistical analysis and correlation analysis are performed on multi-dimensional data to identify potential influencing factors of damage and generate a damage development trend model.

4. The method for predicting apparent damage trend of existing buildings according to claim 1, characterized in that: In step S4, the 4D network model combines time T, space X, Y, and feature dimension F to predict damage trends. The 4D network model consists of the following parts: Input layer: Receives multi-dimensional data, including sensor data, environmental factors, and historical damage records; Convolutional layer: used to extract spatial features and identify damage patterns; Circular layer: captures time series data and identifies damage trends over time; Attention mechanism: Enhances the model's focus on key damage features and improves prediction accuracy; Output layer: outputs the prediction results of future damage status.

5. The method for predicting apparent damage trend of existing buildings according to claim 4, characterized in that: The construction of the 4D network model includes the following processes: Set up a 4D tensor ,in, For time, is the spatial coordinate X axis, is the spatial coordinate Y axis, is the feature dimension; The formula of the 4D network model is as follows: ； in, is the output of the convolutional layer, is the convolution kernel, and its size is set according to actual needs. is the bias term; Use recurrent layers to process time series data: ； in, is the output of the recurrent layer, indicating the change of damage characteristics over time.

6. The method for predicting apparent damage trend of existing buildings according to claim 4, characterized in that: The attention mechanism introduced by the 4D network model is as follows: ； in, is the attention weight, is the learned weight matrix, is the feature dimension; Finally, the prediction result of the output layer is expressed as: ； in, For damage status prediction.

7. The method for predicting apparent damage trend of existing buildings according to claim 1, characterized in that: In step S5, cross-validation is used to evaluate the performance of the model, including the following specific processes: Data segmentation: Divide the preprocessed dataset into training set, validation set and test set; Model training: Use the training set to train the 4D network model; Cross-validation: Use k-fold cross-validation, including k=5, to divide the training set into k subsets, and perform k training cycles, each time using one subset as the validation set and the other subsets as the training set; Performance evaluation: Calculate the performance indicators of the model through the validation set, including root mean square error and accuracy; Parameter tuning: adjust model parameters based on evaluation results; Final training: After multiple rounds of training and tuning, the entire training set is trained using the finalized hyperparameters.

8. The method for predicting apparent damage trend of existing buildings according to claim 7, characterized in that: In the model training, the loss function in the training process selects mean square error or cross entropy loss, and the formula is as follows: ； in, For loss, is the sample size, is the true value, is the predicted value; The parameter tuning includes: Learning rate: Use learning rate scheduling strategies, including cosine annealing, to dynamically adjust the learning rate; Batch size: Try different batch sizes to optimize training speed and stability; Regularization: Add L1 or L2 regularization terms to prevent the model from overfitting.

9. The method for predicting apparent damage trend of existing buildings according to claim 1, characterized in that: In step S6, the dynamic association mechanism includes the following specific processes when it is constructed: Real-time data input: real-time monitoring data input, including sensor data, environmental factors and historical damage records; State update: When new data is input, the model state is updated and the new damage state is calculated using the following formula: : ； in, is the damage status at the current moment, is the damage status at the previous moment, is the damage change calculated based on the new input data; Feedback loop: Compare the model's prediction results with the actual monitoring data and calculate the prediction error. The calculation formula is as follows: ； in, is the prediction error, is the actual monitoring value, is the model prediction value; Monitoring plan adjustment: Dynamically adjust the monitoring plan based on the prediction error. If the error exceeds the set threshold, increase the sensor collection frequency or change the monitoring point to obtain more accurate data; Regular model retraining: Set up regular model retraining to update model weights using the latest monitoring data to maintain accuracy and effectiveness.

10. The method for predicting apparent damage trend of existing buildings according to claim 1, characterized in that: In step S7, the prediction results are compared with the actual monitoring results to analyze the accuracy and recall rate of the model to ensure the reliability of the prediction results of the model; Long-term tracking and monitoring will be carried out to comprehensively evaluate the prediction results of the model in different time periods.

Images