A method for detecting the stability of cave temples based on the periodic characteristics of freeze-thaw cycles

Abstract

The present invention discloses a method for detecting the stability of grotto temples based on the periodic characteristics of freeze-thaw cycles. The method includes: constructing a grotto temple microstructural degradation model from a microscopic perspective to explore the effects of freeze-thaw cycles and dry-wet cycles on microscopic particles; from a local perspective, using a local key crack cracking model to extract the similarity of subsequences between cycles, and designing a time series decomposition module. Multi-layer convolution operations are used to capture short-term temporal dependencies between data, and RNN operations are used to capture long-term temporal dependencies between data. A self-attention module is used to consider the intensity of the effects of different variables; based on the damage limit state equation, the overall stability of the grotto temple is predicted in combination with the cycle, service life, material, and geographical location, thereby improving the safety of the grotto temple. The present invention predicts the local crack development trend and overall stability of the grotto temple in the future, facilitating the early development of response plans and the implementation of protective measures, thereby improving the stability and safety of the grotto temple.

Description

Technical Field

[0001] The present invention relates to the field of rock stability detection in geotechnical engineering, and in particular to a cave temple stability detection method based on the periodic characteristics of freeze-thaw cycles. By processing and learning external environmental data and cave temple crack width data collected over a period of time, the crack development trend of the cave temple in the future over a long period of time is analyzed, and then early warning information is issued in advance, thereby effectively protecting the cave temple. Background Art

[0002] Grotto temples are a perfect fusion of architectural art and religious culture, imbued with profound historical heritage, cultural background, and artistic value. Over thousands of years of history, they have suffered severe damage from both natural and human influences, including cracks, instability, and water seepage. Research has found that weathering primarily influences these damages, with alternating temperature swings, wet-dry cycles, and freeze-thaw cycles as key factors. Wet-dry cycles weaken the connections between mineral particles within the rock, significantly affecting its structure and causing alternating expansion of clay minerals, resulting in degradation. Freeze-thaw cycles involve the frequent freezing and thawing of water within the rock due to temperature fluctuations. This freeze-thaw cycle can disrupt the rock's internal structure and damage it. The protection of grotto temples began in my country in the early days of the People's Republic of China. Initially, simple support measures were implemented. With the development of social and economic science, engineering methods were used to strengthen and protect them. Today, grotto temples are protected in collaboration with other disciplines, indicating that more and more fields are paying more attention to cultural relic protection. However, there is currently a lack of a method that can comprehensively utilize various prior knowledge and use reasonable means to predict the future development trend of grotto temple cracks.

[0003] A time series is a set of observations formed by arranging statistical values in chronological order. It can generally be described as consisting of four components: long-term trends, cyclical fluctuations, seasonal fluctuations, and irregular fluctuations. A long-term trend refers to an overall trend of change influenced by a certain factor over a long period of time. Cyclic fluctuations refer to periodic fluctuations in data over a period of several years. Seasonal fluctuations refer to periodic fluctuations within a year caused by the changing seasons. Irregular fluctuations refer to irregular fluctuations. Time series forecasting can be described as analyzing the dynamics of historical data to uncover underlying patterns and, based on these patterns, predict future values. Today, time series forecasting is widely used in various fields, such as finance, where it can be used to forecast market sales; agriculture, where it can be used to predict annual crop yields; energy, where it can be used to predict and assess various types of energy consumption in factories; and the environment, where it can be used for weather forecasts and PM2.5 alerts. As a key area for identifying future trends, time series forecasting has attracted considerable attention from researchers both domestically and internationally. Forecasting methods can generally be categorized into traditional statistical methods and deep learning methods. Statistical models include naive estimation models, autoregressive integrated moving average (ARIMA) models, exponential smoothing models, and multiple regression models. These models have achieved significant progress in solving time series forecasting problems. However, as their application areas have broadened, time series data has become more complex, facing limitations such as nonlinearity, data sparsity, and a greater number of potential dependencies between data. This makes it difficult for traditional statistical methods to easily extract data features and produce stable and accurate forecasts. Therefore, machine learning methods have been introduced, utilizing models such as RNNs, LSTMs, GRUs, and Auto-Encoders to address more complex time series forecasting problems. AutoFormer combines deep learning with statistical methods. By embedding sequence decomposition units into a deep model, it achieves progressive forecasting and extracts trend information from latent variables in trend terms, resulting in a more robust model and improved long-term time series forecasting. However, these methods have not fully exploited the similarities between time series data periods and the correlations between data variables. Furthermore, current models lack reasonable interpretability. Summary of the Invention

[0004] The present invention provides a method for detecting the stability of grotto temples based on the periodic characteristics of freeze-thaw cycles. The method predicts the development trend of local cracks and the overall stability of grotto temples over a period of time in the future, facilitating the relevant staff to study response plans and take protective measures in advance, thereby improving the stability and safety of grotto temples. Detailed description is provided below:

[0005] A method for detecting the stability of a cave temple based on the periodic characteristics of freeze-thaw cycles, the method comprising:

[0006] From a microscopic perspective, a microstructural degradation model of grotto temples was constructed to explore the effects of freeze-thaw cycles and dry-wet cycles on microparticles.

[0007] From a local perspective, we use the local key cracking model to extract the similarity of subsequences between cycles and design a time series decomposition module. We use multi-layer convolution operations to capture short-term temporal dependencies between data and RNN operations to capture long-term temporal dependencies between data. We also use a self-attention module to consider the influence of different variables.

[0008] Based on the damage limit state equation, the overall stability of the grotto temple is predicted in combination with the cycle, service life, material and geographical location, which improves the safety of the grotto temple.

[0009] The microstructure degradation model of the grotto temple includes:

[0010] The data from different grottoes and temples were preprocessed to create a new dataset through data augmentation, standard deviation denoising, and missing value parameterization.

[0011] Clustering was done on geographical location features and external environmental features. By examining the damage to the grottoes under different clustering conditions, the conclusion on the influence of external physical effects on the distribution of microscopic particles in the rock mass was drawn.

[0012] Furthermore, the time series decomposition module is expressed as:

[0013] Y t =S t +T t +R t t=1,...,T

[0014] Among them, S t is the periodic component in the time series data, T t is the trend component in time series data, R t is the residual component in the time series data.

[0015] Among them, the self-attention module is: adding feature weights to different environmental variable data to perform weighted operations to obtain the dependency between different time series data, combining the dependency of the same time series data in the time dimension with the dependency of variables between different time series data, and obtaining the comprehensive influence of the external environment on local stability.

[0016] Furthermore, the damage limit state equation is:

[0017]

[0018] Where Δ represents the critical fatigue damage, N drepresents the number of freeze-thaw cycles, n is the external action years, e represents the material factor coefficient, K represents the geographical location factor. According to this equation, the local crack changes are predicted. S eq is the comprehensive impact of the external environment on local stability, and D is the overall damage to the grotto temple.

[0019] The beneficial effects of the technical solution provided by the present invention are: using prior knowledge in the field of geotechnical mechanics to design a network architecture, enhancing the interpretability of the model, improving model performance, and predicting the stability of grottoes and temples in the future, so that experts in related fields can take preventive protection measures for grottoes and temples in a timely manner. This solution not only achieves certain theoretical breakthroughs in the field of grotto and temple protection, but also, as a new idea, can provide assistance for other cultural relics protection work in my country. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of a cave temple stability detection method based on the periodic characteristics of freeze-thaw cycles;

[0021] Figure 2 This is the structural diagram of the key network model;

[0022] Figure 3 It is a decomposition diagram of univariate time series data;

[0023] Figure 4 It is a decomposition diagram of multivariate time series data;

[0024] Figure 5 A diagram of the data collection environment;

[0025] Figure 6 Visualize the model prediction results;

[0026] Figure 7 This is a line chart comparing the location of grotto temple materials;

[0027] Figure 8 This is a bar chart comparing the locations of materials in the grotto temples. DETAILED DESCRIPTION

[0028] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention are described in further detail below.

[0029] Time series prediction has been widely used in many fields and has achieved good results. However, its application in geotechnical engineering is almost blank. This is because the external environment of grottoes and temples is complex, and the collected data changes slightly and has many anomalies, making it difficult to extract data features and discover data patterns. Based on this, the factors affecting the stability of grottoes and temples largely come from the external environment, among which the temperature, humidity, rainfall and other data in the external environment can be represented as time series. Then, with the help of the knowledge of time series prediction in deep learning, the intensity of weathering that the grottoes and temples will suffer in the future and the development of rock degradation cracks can be predicted, thereby analyzing the stability of the grottoes and temples as a whole.

[0030] Based on this, the present invention established a data collection point at the Northern Grottoes Temple in Gansu Province, denoised the collected data, and created its own dataset. Furthermore, the present invention innovatively combined knowledge of freeze-thaw cycles and wet-dry cycles in geotechnical engineering with deep learning, attempting to leverage prior theoretical knowledge to inform the model and design an interpretable network architecture. Ultimately, a grotto temple stability prediction model based on the periodic characteristics of freeze-thaw cycles was proposed. This model helps predict the development trends of local cracks and the overall stability of the grotto temples over the coming period, facilitating the early development of response plans and the implementation of protective measures by relevant personnel.

[0031] The embodiment of the present invention mainly describes a network model for long-term prediction of the stability of cave temple rock mass based on the periodic characteristics of freeze-thaw cycles. The overall framework of the proposed method is as follows: Figure 1 As shown, the network model is Figure 2 shown.

[0032] The entire model consists of a grotto temple microstructure degradation model, a local key crack degradation model, and an overall degradation model. Starting from a microscopic perspective, the grotto temple microstructure degradation model is used to explore the effects of physical effects such as freeze-thaw cycles and dry-wet cycles on microscopic particles, explore their long-term laws, and provide a theoretical basis for subsequent work. After understanding the laws governing the effects of freeze-thaw cycles on rock microstructures, the local key cracking model is used to extract the similarity of subsequences between cycles from a local perspective. In this process, a time series decomposition module is designed to decompose the time series data into several potential data patterns. Multi-layer convolution operations are then used to capture short-term temporal dependencies between data, and RNN operations are used to capture long-term temporal dependencies between data. At the same time, a self-attention module is used to consider the strength of the effects of different variables. Finally, the overall stability of the grotto temple is predicted in the overall degradation model using the damage limit state equation proposed in an embodiment of the present invention.

[0033] 1. Microstructural Deterioration Model of Grotto Temples

[0034] Considering that the main cause of damage to the grotto temples is weathering, of which freeze-thaw cycles and dry-wet cycles are the main factors causing weathering, this is essentially due to the continuous change of water content in the rock during the freeze-thaw cycles and dry-wet cycles, which causes the internal particle structure of the rock to continuously shrink and expand. The external manifestation is the appearance of cracks on the surface of the rock mass, and the phenomenon of exfoliation and corrosion. Therefore, in order to explore the impact of this microscopic degradation, the embodiment of the present invention first preprocesses the data from different grotto temples, creates a new data set through data enhancement, standard deviation denoising, and missing value parameterization, and then performs cluster analysis on different samples, including: geographical location feature clustering, external environment feature clustering, and by observing the damage of the grotto temples under different clustering conditions, the conclusion that external physical effects affect the distribution of microscopic particles in the rock mass is confirmed. This part provides a good theoretical basis for the experiments subsequent to the embodiment of the present invention.

[0035] 2. Deterioration model of local key cracks in grotto temples

[0036] In view of the complex external environment of the grotto temples and the numerous dependencies between data, the embodiment of the present invention designs a time series decomposition module that uses the STL (robust locally weighted regression as a smoothing method) time series decomposition method to decompose the original time series into several potential patterns, including periodic factors, trend factors, and residuals. The impact of each potential pattern is then analyzed in a targeted manner, and finally integrated to obtain more complete data information. The time series can be expressed as:

[0037] Y t =S t +T t +R t t=1,...,T (1)

[0038] Through the results of the time series decomposition module, it is found that the data has obvious periodic characteristics. The decomposition results are as follows: Figure 2 Figure 3 As shown, considering that freeze-thaw cycles are an important factor affecting the damage to grotto temples and also have periodic characteristics, the periodic laws of freeze-thaw cycles can help the embodiments of the present invention to well explore the changing laws of grotto temple crack data. Therefore, when designing the capture of short-term time-dependent features and long-term time-dependent features of data, the embodiments of the present invention use the period of freeze-thaw cycles as an important reference factor. For example: when the embodiments of the present invention use convolution (CNN) operations to capture local dependencies within a short-cycle time dimension, since the size k of the convolution kernel reflects how large a range of states a unit state is caused by, the selection of the convolution kernel size of the first layer of convolution refers to the period of the freeze-thaw cycle phenomenon within a day, and the selection of the convolution kernel size of the second layer of convolution refers to the period of the freeze-thaw cycle phenomenon within a week.

[0039] When using a recurrent neural network (RNN) to capture dependencies in the long-term time dimension, if one wants to predict the stability of a cave temple at time t, one can use the data records at time t within the historical period, rather than the data records at adjacent times t. Therefore, the embodiment of the present invention uses the special RNN LSTM and designs a period value, period, to consider data points at the same time belonging to different periods within a time window. In other words, only the unit states at time t and time t-period are considered, skipping the period hidden units in the middle. The selection of the period value is based on the period (one quarter) of the freeze-thaw cycle phenomenon that occurs over a longer period of time, as known in the embodiment of the present invention. This method makes the parameter selection of the neural network highly interpretable. Short-term and long-term time dependencies are then integrated to obtain a complete dependency relationship of the data in the time dimension.

[0040] In addition, many studies have shown that the impact of temperature changes on rock weathering is different from the impact of humidity changes on rock weathering. Therefore, the embodiment of the present invention introduces a self-attention mechanism, which considers the damage factors caused by different external environmental factors to the grotto temples, adds feature weights to different environmental variable data, and performs weighted operations to obtain the dependency relationship between different time series data. Finally, the dependency relationship of the same time series data in the time dimension and the dependency relationship between variables of different time series data are combined to obtain the comprehensive impact of the external environment on local stability.

[0041] 3. Overall Deterioration Model of Grotto Temples

[0042] The comprehensive influence of the external environment on the local stability obtained in the local model is described as S eq , the embodiment of the present invention summarizes the damage limit state equation of the overall stability of the cave temple as:

[0043]

[0044] Where Δ represents the critical fatigue damage, N d Represents the number of cycles of freeze-thaw cycles, n is the external action years, e represents the material factor coefficient, and K represents the geographical location factor. Based on this equation, we predict the changes in local cracks and analyze the local stability. It can be found that the function curves of grottoes in different geographical locations and belonging to different materials are different, which means that they are affected differently by the freeze-thaw cycle, resulting in different development of cracks and damage to the grottoes. At the same time, we take into account that when multiple local areas are affected, it will rise to the effect on the whole. Therefore, the local information obtained is integrated and reasonably inferred to derive the overall stability prediction model of the grottoes.

[0045] Experimental setup

[0046] The experimental code model of the embodiment of the present invention is built based on Python3.6 and Pytorch framework, using the Adam optimizer, setting the initial learning rate to 0.01, the batch_size to 32, and the prediction time window to 12*0.5, 12*1, 12*2, 12*15, and 12*30, that is, the length of a single prediction is gradually extended from a short-term half a day to a long-term one month. In the local crack prediction module of the embodiment of the present invention, there is a short-period time series prediction module, in which the convolution kernel is selected as 12*1, a long-period time series prediction module, in which the period is selected as 12*30*3, and a variable contribution factor analysis module.

[0047] 1. Dataset and evaluation metrics

[0048] (1) Dataset: The Beishiku Temple in Qingyang City, Gansu Province, was selected as the data collection site to monitor the external environment of the temple and the rock cracks. The sampling frequency of the external environment data was 20 minutes, and the sampling frequency of the crack width data was 2 hours. Finally, after data cleaning and denoising, a new complete dataset was created. This dataset contains the temperature, humidity, frost point inside the caves of the temple and the crack width outside the temple from October 2020 to February 2022.

[0049] During the experiment, the dataset was divided into a training set (60%), a validation set (20%), and a test set (20%).

[0050] Evaluation indicators: The embodiment of the present invention uses two common evaluation indicators to evaluate the effectiveness of the model: root mean square error (RMSE) and mean absolute error (MAE). The lower the values of RMSE and MAE, the more accurate the model prediction effect.

[0051] 2. Experimental performance comparison

[0052] The present invention evaluated the proposed model and compared it with advanced time series data prediction methods, including convolutional neural networks (CNN), long short-term memory networks (LSTM), a variant of long short-term neural networks (ConvLSTM), and long short-term temporal pattern modeling based on deep neural networks (LSTNet). Table 1 shows the comparative experimental results of the five methods. It can be seen that the cave temple stability prediction model (Ftcp) based on the periodic characteristics of freeze-thaw cycles proposed in the present invention performed better than other models on the self-built data set and achieved good data prediction results. The prediction results of the model are visualized as shown in the figure below. Figure 5 shown.

[0053] Table 1 Comparative experimental results of five methods under different prediction windows

[0054]

[0055] To ensure that each module in the network architecture proposed in this embodiment of the present invention positively impacts model prediction performance, we conducted ablation experiments, removing one module at a time from the network framework. These experiments included: removing the data decomposition component (Ftcp / decompose), removing the convolutional component (Ftcp / cnn), removing the LSTM component containing periodic values (Ftcp / Lstm-p), and removing the self-attention component from the complete network module. The experimental results are shown in Table 2.

[0056] Table 2 Ablation experiment results

[0057]

[0058]

[0059] The crack and fissure conditions of the grotto temples were evaluated using the formula proposed in the embodiment of the present invention. It was found that the damage to the grotto temples was related to their geographical location (k) and the material (e). For example, Figure 7 Figure 8 shown.

[0060] According to Table 1, Table 2, Figure 7 、 Figure 8 The following conclusions can be drawn:

[0061] The model of the embodiment of the present invention has the best prediction performance compared with all the comparison methods. In particular, when performing long-term data prediction, the model of the embodiment of the present invention is more stable and the performance has not been significantly reduced compared with the comparison methods. This shows that the model of the embodiment of the present invention can not only capture short-term time series data dependencies, but also shows great potential in capturing long-term time series data dependencies, which can meet the purpose of the embodiment of the present invention to perform long-term prediction under the premise of only having data values within a shorter time range.

[0062] Ablation experiments have proved that each component in the model of the embodiment of the present invention is extremely important, and the combined effect of each component makes the model have good robustness. It can be found that when the LSTM component with the period value Period is removed, the model effect decreases most significantly, indicating that when making long time series predictions, capturing the similarity of time series data within a long period range is crucial in the model of the embodiment of the present invention.

[0063] The formula proposed in the embodiment of the present invention proves that the degree to which the grotto temple is affected by the external environment such as freeze-thaw cycles is also closely related to the material of the grotto temple and the geographical location, which should also be taken into consideration to obtain a more comprehensive result.

[0064] The embodiments of the present invention have experimentally demonstrated that neural networks can be designed based on prior knowledge. The interpretability of neural networks has always been a concern for many researchers. The model of the embodiments of the present invention provides a solution, namely, using existing theoretical knowledge to explore data patterns, combining theoretical knowledge with neural network knowledge, and designing the network architecture more logically.

[0065] In general, the present invention proposes a model for predicting the stability of cave temples based on the periodic characteristics of freeze-thaw cycles. This model makes full use of existing prior knowledge and combines the knowledge of time series prediction in neural networks with the geotechnical stability theory in geotechnical engineering to predict the overall stability of cave temples, making it easier for relevant staff to study response plans in advance and take protective measures. The core of the model (Ftcp) of the present invention lies in capturing the similarity of long- and short-period time series data subsequences respectively. Based on the knowledge of freeze-thaw cycles, a short-term time dependency capture module and a long-term time dependency capture module are designed to extract data features in the time dimension. At the same time, the variable dependency captured by the self-attention module significantly improves the accuracy of multivariate time series data prediction and enhances the stability of multivariate time series data prediction over a longer range. Comprehensive experimental results show that this method can accurately predict the stability trend of cave temples. It can be said that the model of the embodiment of the present invention has reached the most advanced level.

[0066] Unless otherwise specified, the embodiments of the present invention do not limit the models of the components. Any component that can perform the above functions may be used.

[0067] Those skilled in the art will understand that the accompanying drawings are only a schematic diagram of a preferred embodiment, and the serial numbers of the embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting the stability of cave temples based on the periodic characteristics of freeze-thaw cycles, characterized in that: The method comprises: From a microscopic perspective, a microstructural degradation model of grotto temples was constructed to explore the effects of freeze-thaw cycles and dry-wet cycles on microparticles. From a local perspective, we use the local key cracking model to extract the similarity of subsequences between cycles and design a time series decomposition module. We use multi-layer convolution operations to capture short-term temporal dependencies between data and RNN operations to capture long-term temporal dependencies between data. We also use a self-attention module to consider the influence of different variables. Based on the damage limit state equation, the overall stability of the grotto temple was predicted by combining the cycle, service life, material and geographical location, thus improving the safety of the grotto temple. The microstructure degradation model of the grotto temple includes: The data from different grottoes and temples were preprocessed to create a new dataset through data augmentation, standard deviation denoising, and missing value parameterization. Clustering geographical location characteristics and external environmental characteristics, by observing the damage of the cave temples under different clustering conditions, as a conclusion on the influence of external physical effects on the distribution of microscopic particles in the rock body; Wherein, the time series decomposition module is expressed as: Y t =S t +T t +R t t=1,...,T Among them, S t is the periodic component in the time series data, T t is the trend component in time series data, R t is the residual component in the time series data; The self-attention module adds feature weights to different environmental variable data to perform weighted operations to obtain the dependencies between different time series data, combines the dependencies of the same time series data in the time dimension with the dependencies of variables between different time series data, and obtains the comprehensive impact of the external environment on local stability. The damage limit state equation is: Where Δ represents the critical fatigue damage, N d represents the number of cycles of freeze-thaw cycles, n is the external action years, e represents the material factor coefficient, K represents the geographical location factor, and based on this equation to predict the local crack changes, S eq is the comprehensive impact of the external environment on local stability, and D is the overall damage to the grotto temple.

Images