A method and device for predicting the tilt of cultural heritage buildings

Abstract

This invention relates to the field of cultural relic building protection and intelligent predictive modeling technology, and provides a method and apparatus for predicting the tilt of cultural relic buildings. In this embodiment, for a target cultural relic building whose tilt needs to be predicted, tilt monitoring data of the building is first collected. Then, the weight values ​​of each indicator in the tilt monitoring data are determined through the building's process data and environmental data. Features are then extracted based on these weight values. Next, based on the aforementioned process data and environmental data, a target model for predicting the tilt of the cultural relic building is determined from several preset models. Finally, the extracted features are used to predict the tilt of the building through this model, thereby determining the tilt result. This application uses different models for buildings with different processes and extracts features from the tilt monitoring data based on the building's process and environment, which significantly improves the accuracy of tilt prediction for cultural relic buildings.

Description

Technical Field

[0001] This application relates to the field of cultural heritage building protection and intelligent prediction modeling technology, and in particular to a method and device for predicting the tilt of cultural heritage buildings. Background Technology

[0002] Heritage buildings, as important material carriers of human history and culture, are non-renewable. Over time, due to factors such as uneven foundation settlement, environmental erosion, disturbance from surrounding construction, and material aging, heritage buildings often suffer structural damage such as tilting and cracking. Among these, tilting is a key indicator reflecting the structural safety status of heritage buildings, and accurate prediction of tilting is of great significance for preventive protection, early warning, and the development of repair plans.

[0003] Currently, the monitoring and prediction of tilting of historical buildings mainly rely on the following types of technical methods: Firstly, regarding the acquisition of monitoring data, existing technologies typically involve deploying tilt sensors, total stations, or static levels at key measuring points of historical buildings to obtain real-time monitoring data such as tilt angles and tilt rates. However, raw monitoring data often contains noise, outliers, or fluctuations caused by environmental interference. If used directly for prediction, it can easily lead to inaccurate prediction results.

[0004] Secondly, in terms of tilt trend prediction, existing methods mostly use a single time series analysis model (such as autoregressive moving average model, grey model) or simple regression analysis, which extrapolates and predicts based solely on historical tilt data. Although such methods can reflect the changing patterns of data over time, there are significant differences between different buildings, making it difficult to guarantee the reliability of the prediction results in many cases. Summary of the Invention

[0005] In view of this, this application provides a method and apparatus for predicting the tilt of cultural heritage buildings, which adjusts the weights and selects models for the tilt data of cultural heritage buildings by using the technological parameters and environmental parameters of the cultural heritage buildings, thereby improving the accuracy of tilt prediction of cultural heritage buildings.

[0006] The first aspect of this application provides a method for predicting the tilt of cultural relics and buildings, the method comprising: Acquire initial tilt monitoring data from several sets of measuring points of the target cultural relic building, and preprocess the initial tilt monitoring data to obtain tilt monitoring data for model prediction. The tilt monitoring data includes tilt angle, tilt rate, and monitoring time. The process data and environmental data of the target cultural relic building are determined. The weight values ​​of each indicator in the tilt monitoring data of each measuring point are determined through the process data and environmental data. The first feature matrix of each measuring point is determined based on the tilt monitoring data and the weight values. The second feature matrix of the target cultural relic building is then determined based on the process data and environmental data. The second feature matrix is ​​used to determine the target tilt prediction model from several preset models. The first feature matrix is ​​input into the target tilt prediction model to obtain the tilt prediction value of each measuring point output by the target tilt prediction model. The tilt prediction result of the target cultural relic building is determined by the tilt prediction value.

[0007] Optionally, the preprocessing of the initial tilt monitoring data includes: For missing data, the estimated value of the missing value is calculated by the least squares method, and the missing value is filled in using the estimated value; For outliers, the mean μ and standard deviation σ of the tilt angle in the tilt monitoring data are determined. Data that exceeds the interval [μ-3σ, μ+3σ] are identified as outliers and removed. The median within this interval is then used to replace the outlier.

[0008] Optionally, after obtaining the tilt prediction result of the target cultural relic building, the method further includes: The initial prediction results are adjusted using the process data and environmental data, and the adjusted tilt prediction results are determined as the final tilt prediction results for the target cultural relic building.

[0009] Optionally, the preset models include the ELM model using the ReLU activation function, the KRR model using the RBF kernel function, and the XGBoost model.

[0010] Optionally, determining the target tilt prediction model from several preset models using the second feature matrix includes: The second feature matrix is ​​input into each preset model to determine the mean absolute percentage error (MAPE) of each preset model, and the preset model with the smallest MAPE is determined as the target tilt prediction model of the target cultural relic.

[0011] A second aspect of this application provides a device for predicting the tilt of historical buildings, the device comprising: The unit comprises a tilt data acquisition unit, a feature matrix determination unit, and a tilt result prediction unit. The tilt data acquisition unit is used to acquire initial tilt monitoring data of several sets of measuring points of the target cultural relic building, and to preprocess the initial tilt monitoring data to obtain tilt monitoring data including tilt angle, tilt rate and monitoring time. Specifically, it includes: for missing data, calculating the estimated value of the missing value by the least squares method, and supplementing the missing value by the estimated value. For outliers, the mean μ and standard deviation σ of the tilt angle in the tilt monitoring data are determined. Data that exceeds the interval [μ-3σ, μ+3σ] are identified as outliers and removed. The median within this interval is used to replace the outliers. The feature matrix determination unit is used to retrieve the process data and environmental data of the target cultural relic building, calculate the weight value of each indicator in the tilt monitoring data of each measuring point based on the process data and environmental data, generate the first feature matrix of each measuring point by combining the weight value and the tilt monitoring data, and construct the second feature matrix of the target cultural relic building based on the process data and environmental data. The tilt prediction unit is used to retrieve several preset models to select and determine the target tilt prediction model. The first feature matrix is ​​input into the target tilt prediction model to obtain the tilt prediction value of each measuring point, and the tilt prediction result of the target cultural relic building is generated based on the tilt prediction value. The several preset models include the ELM model with ReLU activation function, the KRR model with RBF kernel function, and the XGBoost model. The selection method is as follows: the second feature matrix is ​​input into each preset model, the mean absolute percentage error (MAPE) of each preset model is determined, and the preset model with the smallest MAPE is determined as the target tilt prediction model of the target cultural relic.

[0012] Optionally, the tilted data acquisition unit includes a data verification module, a missing value completion module, and an outlier correction module; The data verification module is used to verify the numerical fields of the initial tilt monitoring data and convert them into float64 type. The missing value completion module is used to calculate the estimated value of the missing value using the least squares method and complete the missing value completion. The outlier correction module is used to identify outliers based on the 3σ criterion, and replace them with the median of the interval after removing them.

[0013] Optionally, the device further includes: The tilt result update unit is used to adjust the initial prediction result based on the process data and environmental data after obtaining the tilt prediction result of the target cultural relic building, and to determine the adjusted tilt prediction result as the final tilt prediction result of the target cultural relic building.

[0014] Optionally, the tilt result prediction unit has a built-in model storage module and a model evaluation module; The model storage module is used to store the ELM model with ReLU activation function, the KRR model with RBF kernel function, and the XGBoost model, and each model has completed pre-training and parameter optimization. The model evaluation module is used to input the second feature matrix into each preset model, calculate and compare the mean absolute percentage error (MAPE) of each model, and output the model with the smallest MAPE as the target tilt prediction model.

[0015] In the embodiments provided in this application, for a target cultural relic building whose tilt needs to be predicted, tilt monitoring data of the building is first collected. Then, the weight values ​​of each indicator in the tilt monitoring data are determined through the building's technological and environmental data. Features are then extracted based on these weight values. Next, based on the aforementioned technological and environmental data, a target model for predicting the tilt of the cultural relic building is determined from several preset models. Finally, the extracted features are used to predict the tilt of the building through this model, thereby determining the tilt result. This application uses different models for buildings with different technological processes and extracts features from the tilt monitoring data based on the building's technological processes and environment, which significantly improves the accuracy of tilt prediction for cultural relic buildings.

[0016] Furthermore, the process and environmental data in this application are not only used for dynamic weight adjustment and model selection of tilt monitoring data, but also for correcting the prediction results during the tilt prediction process, thereby further improving the accuracy of the prediction. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the method provided in this application embodiment; Figure 2 Comparison chart of model training results provided in the embodiments of this application; Figure 3 Another comparison chart of model training results provided in this application embodiment; Figure 4 This is a measurement point-predicted value mapping diagram provided in the embodiments of this application; Figure 5 This is a structural diagram of the device provided in the embodiments of this application; Figure 6 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0018] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0019] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0020] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0021] This application provides a method and apparatus for predicting the tilt of cultural heritage buildings, so as to improve the accuracy of tilt prediction for cultural heritage buildings.

[0022] The technical solutions of this application will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0023] like Figure 1 The diagram shown is a flowchart of a method for predicting the tilt of a historical building provided in this application. The process may include the following steps: Step S101: Obtain initial tilt monitoring data of several sets of measuring points of the target cultural relic building, and preprocess the initial tilt monitoring data to obtain tilt monitoring data for model prediction.

[0024] In this embodiment, the tilt monitoring data includes the tilt angle, tilt rate, monitoring time, etc., of the historical building. The tilt angle can be measured using a wireless tilt sensor, employing MEMS capacitive or electrolyte sensing elements to calculate the tilt angle by measuring the component of gravitational acceleration along the sensitive axis. The tilt rate is calculated differentially from the tilt angle data, by calculating the angle change per unit time based on continuously collected tilt angle values ​​at time intervals. The initial tilt monitoring data may also include the measurement point number for each measuring point; this number is a unique identifier used to identify the tilt monitoring data at each location. If the initial tilt monitoring data does not contain this unique identifier, it needs to be added during post-processing.

[0025] After data collection, numerical fields are validated and converted to float64 type. For missing values, linear interpolation is used. A linear fitting equation y=kx+b is constructed based on three valid data points before and after the missing data. The parameters of the equation are solved using the least squares method to calculate the estimated value of the missing value. Outliers are identified based on the 3σ criterion. The mean μ and standard deviation σ of the tilt angle data in the training set are calculated first. Data outside the interval [μ-3σ, μ+3σ] are identified as outliers, removed, and replaced with the median within that interval.

[0026] Finally, the sorted monitoring data was divided into training and test sets in an 8:2 ratio, with the first 80% of the data used as the training set and the last 20% used as the test set, laying the foundation for subsequent model training and evaluation.

[0027] Step S102: Determine the process data and environmental data of the target cultural relic building; determine the weight value of each indicator in the tilt monitoring data of each measuring point through the process data and environmental data; determine the first feature matrix of each measuring point based on the tilt monitoring data and weight values; and determine the second feature matrix of the target cultural relic building based on the process data and environmental data.

[0028] In this embodiment, the process data includes information such as the building's structural form, materials, and construction techniques. Structural forms include timber structures, brick and stone structures, brick-timber hybrid structures, and rammed earth structures; materials include timber, red bricks, blue bricks, and the strength of the bricks and stones; construction techniques include mortise and tenon joints, masonry methods, mortar joint thickness, and wall construction. This information can be determined by consulting building archives, historical drawings, and on-site observation to identify the structural system.

[0029] Environmental data can include information such as temperature and humidity, rainfall, soil geology, and precipitation. It can be determined using various types of sensors.

[0030] In this embodiment, a process-weight mapping rule base can be pre-defined. This rule base records the weight vectors of tilt angle, tilt rate, and monitoring time under various conditions. For example, masonry structures are more dangerous due to cumulative deformation, and therefore are more sensitive to angle. The weight value of the tilt angle can be increased. Wooden structures, due to rapid wood creep and temperature / humidity response, are more sensitive to rate, and the weight value of the tilt rate can be increased. Environmental data can be used to adjust the above rules appropriately. For example, in high-humidity environments, since humidity affects the material creep rate, the weight of the tilt rate can be increased.

[0031] Alternatively, a weight determination model can be pre-trained. This model is based on the aforementioned weight allocation strategy and historical data of various cultural relics and buildings. This model can determine the optimal weight allocation under various complex conditions.

[0032] After determining the weight values, features are extracted from the tilt monitoring data, process data, and environmental data respectively. Then, the features of each indicator in the tilt monitoring data are multiplied by their corresponding weight values ​​to form a weighted feature vector. The weighted feature vectors of all monitoring points are then concatenated in order of monitoring point number to construct the first feature matrix. The process data and environmental data are then combined to form the second feature matrix.

[0033] Furthermore, during the construction of the first feature matrix, time-series historical features can be added, covering derived features such as tilt angle values ​​lagged by 1-5 days, tilt rate values ​​lagged by 1-3 days, 5-day rolling mean, 3-day rolling standard deviation, and first-order difference.

[0034] Furthermore, data from different measuring points can be combined into correlation features. For example, does the tilt of a column cause the displacement of a beam? Is the outward tilt of a wall related to the settlement of the foundation? These correlation features reflect the integrity of the structure.

[0035] Step S103: Determine the target tilt prediction model from several preset models using the second feature matrix, input the first feature matrix into the target tilt prediction model to obtain the tilt prediction values ​​of each measuring point output by the target tilt prediction model, and determine the tilt prediction result of the target cultural relic building using the tilt prediction values.

[0036] Because artifacts made using different techniques exhibit completely different deformation patterns, different prediction models are required.

[0037] From a construction technique perspective—for example, even within the same timber structure, the force transmission paths of the northern raised-beam style and the southern mortise-and-tenon style are completely different. The former uses a beam-column system, while the latter uses a column-beam system, which determines their different response modes to deformation. Another example is masonry techniques: the rammed earth walls of the north are strong but susceptible to water damage, while the hollow-beam walls of the south are lightweight but less robust, naturally resulting in different deformation characteristics.

[0038] At the level of material properties—traditional materials from different regions have unique "characters": the aging of blue bricks is slow and logarithmic, the aging of raw soil is power-law shaped, and the aging of wood accelerates after 200 years. These time-varying characteristics of materials determine the development pattern of deformation.

[0039] For cultural relics that deform slowly and exhibit strong regularity, such as official buildings in the north, time-series models that can capture long-term trends are suitable; for cultural relics that deform in a complex manner and are greatly affected by the environment, such as wooden structures in the south, nonlinear models that can handle multiple factors are suitable; for cultural relics with numerous components and complex relationships, such as large palaces, graph network models that can depict the relationships are suitable; and for cultural relics with scarce data but clear physical laws, physical information models that incorporate mechanical equations are suitable.

[0040] Therefore, in this embodiment, several different models can be pre-trained, such as a humidity-sensitive time-series model, whose tilt is significantly affected by humidity and needs to consider periodic fluctuations; a trend extrapolation model, which exhibits strong deformation regularity and is suitable for long-term trend prediction; and a multi-factor coupling model incorporating parameters such as typhoons and rainfall, which predicts tilt conditions under abnormal weather conditions. After training, the predicted tilt values ​​can be compared and verified with the actual values ​​in the test set, and the results are as follows. Figure 2 , Figure 3 As shown, Figure 2 This is a comparison chart of the actual values ​​on the training set and the predicted values ​​on the test set for the three-story south-facing archway west side tilt data using the ELM model. Figure 3 This is a comparison chart showing the actual values ​​from the training set and the predicted results from the test set for the three-story west-facing archway's south-side tilt data using the KRR model. Figure 2 , 3 In the diagram, the blue line represents the true values ​​in the training set, the green line represents the predicted values ​​in the training set, the red line represents the true values ​​in the test set, the yellow line represents the predicted values ​​in the test set, and the gray dashed line is the boundary between the test set and the training set.

[0041] In another embodiment, the aforementioned preset models include the ELM model, the KRR model, and the XGBoost model. The ELM model uses the ReLU activation function, maintains a match between the input layer dimension and the feature matrix dimension, empirically optimizes the number of neurons in the hidden layer, randomly initializes the weights and biases of the input and hidden layers, solves for the output layer weights using the least squares method, and sets a reasonable number of iterations. The KRR model selects the RBF kernel function, optimizes the model parameters through 5-fold time-series cross-validation, and sets a regularization term to balance the model's fitting effect and generalization ability. The XGBoost model uses gradient boosting to construct an ensemble learning framework, controls model complexity by setting tree depth, learning rate, and regularization parameters, and uses the squared error loss function to gradually reduce training error.

[0042] In this embodiment, the target model can be determined using a model matcher. This model matcher employs a random forest classifier or a neural network classifier, pre-trained using the historical second feature matrices of multiple historical buildings and their corresponding optimal prediction model labels as training samples. In practical applications, the second feature matrix of the target historical building at the current moment is input into the model matcher, which outputs a target tilt prediction model type identifier that best matches the current technological and environmental state of the building. The corresponding target tilt prediction model is then determined using this identifier.

[0043] In another embodiment, the second feature matrix can be input into each preset model to determine the mean absolute percentage error (MAPE) of each preset model, and the preset model with the smallest MAPE is determined as the target tilt prediction model for the target cultural relic. This embodiment uses MAPE as the sole evaluation metric to calculate the prediction error of each model on the test set. During the calculation process, samples with an absolute value less than 1e-10 are removed to avoid calculation anomalies, and the training time of each model (accurate to milliseconds) is recorded simultaneously. For each monitoring point, the model with the smallest MAPE value is selected as the optimal model for that monitoring point.

[0044] For example, at the measurement point "west side of the south-facing archway on the third floor", the MAPE of the ELM model is 0.20%, the MAPE of the KRR model is 0.28%, and the MAPE of the XGBoost model is 0.62%. Therefore, the ELM model is selected as the optimal model for this measurement point. At the measurement point "south side of the west-facing archway on the third floor", the MAPE of the ELM model is 0.18%, the MAPE of the KRR model is 0.15%, and the MAPE of the XGBoost model is 0.26%. Therefore, the KRR model is selected as the optimal model for this measurement point.

[0045] After determining the target model, the first feature matrix mentioned above is input into the model, and the predicted tilt values ​​of each measuring point of the target cultural relic are output. Finally, the predicted tilt values ​​of each measuring point are associated with the measuring point number and spatial location to construct a measuring point-predicted value mapping table, such as... Figure 4 As shown, based on historical and predicted tilt data for each measuring point, a tilt trend curve is generated for each measuring point. The predicted tilt angle of each measuring point is compared with a preset safety threshold; when the predicted tilt angle of any measuring point exceeds the warning threshold, a warning message is generated for the corresponding measuring point. Simultaneously, the predicted tilt values ​​of each measuring point are spatially interpolated to generate an overall tilt distribution cloud map of the target historical building, visually displaying the predicted tilt state of various parts of the building. Combining the process data and environmental data of each measuring point, the physical mechanism of the tilt prediction results is explained, and a tilt prediction report is output. This report includes the predicted tilt angle, predicted tilt rate, confidence interval, warning level, and tilt cause analysis based on process-environment characteristics for each measuring point. Through the above methods, the discrete predicted values ​​of each measuring point are transformed into overall building tilt prediction results with spatial distribution characteristics and risk level assessment, providing a decision-making basis for the preventive protection of historical buildings.

[0046] This concludes the process. Figure 1 The process is shown below.

[0047] In this embodiment, for a target cultural relic building whose tilt needs to be predicted, tilt monitoring data of the building is first collected. Then, the weight values ​​of each indicator in the tilt monitoring data are determined through the building's technological and environmental data. Features are then extracted based on these weight values. Next, based on the aforementioned technological and environmental data, a target model for predicting the tilt of the cultural relic building is determined from several preset models. Finally, the extracted features are used to predict the tilt of the building using this model, thereby determining the tilt result. This application uses different models for buildings with different technological processes and extracts features from the tilt monitoring data based on the building's technological processes and environment, which significantly improves the accuracy of tilt prediction for cultural relic buildings.

[0048] In another embodiment, after obtaining the tilt prediction result of the target cultural relic building, the method further includes: The initial prediction results are adjusted using the process data and environmental data, and the adjusted tilt prediction results are determined as the final tilt prediction results for the target cultural relic building.

[0049] This embodiment uses process and environmental data to make the tilt monitoring data prediction tilt process a dynamic process.

[0050] For example, in the first round of prediction: the model outputs preliminary results based on the tilted data.

[0051] The second round of revisions involved incorporating process environment data to adjust forecasts based on current environmental conditions. "We just experienced a minor earthquake; the deformation rate may increase over the next three months, so the forecast curve has been adjusted."

[0052] The third round of verification: The "rules of thumb" provided by the process environment data are used to check the reasonableness of the predictions. For example: "Traditional craftsmanship experience shows that the risk increases sharply when the tilt of this process exceeds 5cm. The current predicted value is 4.8cm, and it is recommended to raise the warning level."

[0053] "This building has never exhibited this deformation pattern under similar environmental conditions in its history, and a re-examination is recommended."

[0054] This embodiment ensures that the prediction results are always under the "supervision" of process environment data, and are continuously calibrated and corrected, thereby further improving the accuracy of the prediction.

[0055] This application also provides a device for predicting the tilt of cultural heritage buildings, such as... Figure 5 As shown, the device includes: The tilt data acquisition unit 501, the feature matrix determination unit 502, and the tilt result prediction unit 503 are included. The tilt data acquisition unit 501 is used to acquire initial tilt monitoring data of several sets of measuring points of the target cultural relic building, preprocess the initial tilt monitoring data to obtain tilt monitoring data including tilt angle, tilt rate and monitoring time, specifically including: for missing data, calculating the estimated value of the missing value by the least squares method, and supplementing the missing value by the estimated value; For outliers, the mean μ and standard deviation σ of the tilt angle in the tilt monitoring data are determined. Data that exceeds the interval [μ-3σ, μ+3σ] are identified as outliers and removed. The median within this interval is used to replace the outliers. The feature matrix determination unit 502 is used to retrieve the process data and environmental data of the target cultural relic building, calculate the weight value of each indicator in the tilt monitoring data of each measuring point based on the process data and environmental data, generate the first feature matrix of each measuring point by combining the weight value and the tilt monitoring data, and construct the second feature matrix of the target cultural relic building based on the process data and environmental data. The tilt prediction unit 503 is used to retrieve several preset models to select and determine the target tilt prediction model, input the first feature matrix into the target tilt prediction model to obtain the tilt prediction value of each measuring point, and generate the tilt prediction result of the target cultural relic building based on the tilt prediction value. The several preset models include the ELM model with ReLU activation function, the KRR model with RBF kernel function, and the XGBoost model. The selection method is as follows: input the second feature matrix into each preset model, determine the mean absolute percentage error (MAPE) of each preset model, and determine the preset model with the smallest MAPE as the target tilt prediction model of the target cultural relic.

[0056] In another embodiment, the tilted data acquisition unit includes a data verification module, a missing value completion module, and an outlier correction module; The data verification module is used to verify the numerical fields of the initial tilt monitoring data and convert them into float64 type. The missing value completion module is used to calculate the estimated value of the missing value using the least squares method and complete the missing value completion. The outlier correction module is used to identify outliers based on the 3σ criterion, and replace them with the median of the interval after removing them.

[0057] In another embodiment, the device further includes: The tilt result update unit 504 is used to adjust the initial prediction result based on the process data and environmental data after obtaining the tilt prediction result of the target cultural relic building, and to determine the adjusted tilt prediction result as the final tilt prediction result of the target cultural relic building.

[0058] In another embodiment, the tilt result prediction unit has a built-in model storage module and a model evaluation module; The model storage module is used to store the ELM model with ReLU activation function, the KRR model with RBF kernel function, and the XGBoost model, and each model has completed pre-training and parameter optimization. The model evaluation module is used to input the second feature matrix into each preset model, calculate and compare the mean absolute percentage error (MAPE) of each model, and output the model with the smallest MAPE as the target tilt prediction model.

[0059] The present invention provides a method for predicting the tilt of cultural relics buildings in the above embodiments, and provides a device for predicting the tilt of cultural relics buildings based on the method. Through the above method and device, the process data, environmental data and tilt data of cultural relics buildings can be linked together, thereby improving the accuracy of tilt prediction of cultural relics buildings.

[0060] This embodiment also discloses a computer device, such as... Figure 6 As shown, the computer device includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to implement any of the above-described methods for predicting the tilt of cultural relics buildings.

[0061] Furthermore, in the above-described implementation of the cultural relic building tilt prediction device, the logical division of each program module is merely illustrative. In actual applications, the above functions can be assigned to different program modules as needed, for example, for the sake of corresponding hardware configuration requirements or the convenience of software implementation. That is, the internal structure of the cultural relic building tilt prediction device can be divided into different program modules to complete all or part of the functions described above.

[0062] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting the tilt of cultural relics and buildings, characterized in that, The method includes: Acquire initial tilt monitoring data from several sets of measuring points of the target cultural relic building, and preprocess the initial tilt monitoring data to obtain tilt monitoring data for model prediction. The tilt monitoring data includes tilt angle, tilt rate, and monitoring time. The process data and environmental data of the target cultural relic building are determined. The weight values ​​of each indicator in the tilt monitoring data of each measuring point are determined through the process data and environmental data. The first feature matrix of each measuring point is determined based on the tilt monitoring data and the weight values. The second feature matrix of the target cultural relic building is then determined based on the process data and environmental data. The second feature matrix is ​​used to determine the target tilt prediction model from several preset models. The first feature matrix is ​​input into the target tilt prediction model to obtain the tilt prediction value of each measuring point output by the target tilt prediction model. The tilt prediction result of the target cultural relic building is determined by the tilt prediction value.

2. The method according to claim 1, characterized in that, The preprocessing of the initial tilt monitoring data includes: For missing data, the estimated value of the missing value is calculated by the least squares method, and the missing value is filled in using the estimated value; For outliers, the mean μ and standard deviation σ of the tilt angle in the tilt monitoring data are determined. Data that exceeds the interval [μ-3σ, μ+3σ] are identified as outliers and removed. The median within this interval is then used to replace the outlier.

3. The method according to claim 1, characterized in that, After obtaining the tilt prediction result of the target cultural relic building, the method further includes: The initial prediction results are adjusted using the process data and environmental data, and the adjusted tilt prediction results are determined as the final tilt prediction results for the target cultural relic building.

4. The method according to claim 1, characterized in that, The preset models include the ELM model using the ReLU activation function, the KRR model using the RBF kernel function, and the XGBoost model.

5. The method according to claim 1, characterized in that, The step of determining the target tilt prediction model from several preset models using the second feature matrix includes: The second feature matrix is ​​input into each preset model to determine the mean absolute percentage error (MAPE) of each preset model, and the preset model with the smallest MAPE is determined as the target tilt prediction model of the target cultural relic.

6. A device for predicting the tilt of cultural relics and buildings, characterized in that, The device includes: The system comprises a tilt data acquisition unit, a feature matrix determination unit, and a tilt result prediction unit. The tilt data acquisition unit is used to acquire initial tilt monitoring data of several sets of measuring points of the target cultural relic building, and to perform the preprocessing operation described in claim 2 on the initial tilt monitoring data to obtain tilt monitoring data including tilt angle, tilt rate, and monitoring time. The feature matrix determination unit is used to retrieve the process data and environmental data of the target cultural relic building, calculate the weight value of each indicator in the tilt monitoring data of each measuring point based on the process data and environmental data, generate the first feature matrix of each measuring point by combining the weight value and the tilt monitoring data, and construct the second feature matrix of the target cultural relic building based on the process data and environmental data. The tilt prediction unit is used to retrieve several preset models as described in claim 4, perform the model filtering operation as described in claim 5 to determine the target tilt prediction model, input the first feature matrix into the target tilt prediction model to obtain the tilt prediction value of each measuring point, and generate the tilt prediction result of the target cultural relic building based on the tilt prediction value.

7. The apparatus according to claim 6, characterized in that... The tilted data acquisition unit includes a data verification module, a missing value completion module, and an outlier correction module. The data verification module is used to verify the numerical fields of the initial tilt monitoring data and convert them into float64 type. The missing value completion module is used to calculate the estimated value of the missing value using the least squares method and complete the missing value completion. The outlier correction module is used to identify outliers based on the 3σ criterion, and replace them with the median of the interval after removing them.

8. The apparatus according to claim 6, characterized in that, The device further includes: The tilt result update unit is used to adjust the initial prediction result based on the process data and environmental data after obtaining the tilt prediction result of the target cultural relic building, and to determine the adjusted tilt prediction result as the final tilt prediction result of the target cultural relic building.

9. The apparatus according to claim 6, characterized in that: The tilt result prediction unit has a built-in model storage module and a model evaluation module; The model storage module is used to store the ELM model with ReLU activation function, the KRR model with RBF kernel function, and the XGBoost model, and each model has completed pre-training and parameter optimization. The model evaluation module is used to input the second feature matrix into each preset model, calculate and compare the mean absolute percentage error (MAPE) of each model, and output the model with the smallest MAPE as the target tilt prediction model.

Images