Foundation pit retaining wall deformation three-dimensional space-time distribution prediction method
The space-time characteristics of foundation pit retaining walls were extracted through the CNN-LSTM-Att model, which solved the problem of predicting retaining wall deformation in foundation pit projects, achieved more accurate deformation prediction and risk assessment, and improved the efficiency and accuracy of project safety management.
Patent Information
- Application Number
- CN202510555745.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-29
- Publication Date
- 2025-08-08
AI Technical Summary
The deformation of the foundation pit retaining wall cannot be predicted during the foundation pit project, resulting in the inability to take timely protective measures, which affects the smooth progress of the project.
The CNN-LSTM-Att model is adopted, combined with convolutional neural network, long-term memory network and attention mechanism, and the spatiotemporal characteristics of the retaining wall are extracted by collecting and preprocessing the foundation pit monitoring data, and the hyperparameters are optimized to predict the three-dimensional spatiotemporal distribution of the retaining wall deformation.
It improves the accuracy and adaptability of retaining wall deformation prediction, enhances the generalization ability and resource utilization efficiency of the model, provides more accurate deformation information and risk assessment, and supports engineering safety management.
Smart Images
Figure CN120449681A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of foundation pit engineering, and in particular relates to a method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall. Background Art
[0002] During the construction of foundation pit projects, the retaining walls of foundation pits often deform. However, engineering managers are unable to predict the deformation of the retaining walls. This results in the inability of engineering managers to take timely and effective protective measures, and thus the inability to ensure the smooth progress of the foundation pit project. Therefore, a three-dimensional spatiotemporal distribution prediction method for the deformation of foundation pit retaining walls is proposed to solve the above problem. Summary of the Invention
[0003] In order to solve the above problems, the present invention provides a method for predicting the three-dimensional spatiotemporal distribution of deformation of foundation pit retaining walls, which solves the problem of being unable to predict the deformation of foundation pit retaining walls.
[0004] The present invention is implemented by the following scheme: A method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall comprises the following steps:
[0005] S1. Collect foundation pit monitoring data and pre-process the monitoring data to obtain a spatiotemporal matrix of retaining wall deformation;
[0006] S2. Construct a CNN-LSTM-Att model, input the spatiotemporal matrix into the model, use a convolutional neural network to extract the spatial features of the retaining wall deformation, use a long short-term memory network to capture the temporal features of the retaining wall deformation, and use an attention mechanism to calculate the attention score of the wall deformation at each time step based on the spatial features and the temporal features;
[0007] S3. Optimize the model’s hyperparameters;
[0008] S4. Based on the optimized model, the deformation of the retaining wall in the future time step is predicted according to the change pattern of the attention score of the wall deformation at each time step.
[0009] A further improvement of the method for predicting the three-dimensional spatiotemporal distribution of foundation pit retaining wall deformation of the present invention is that the foundation pit monitoring data includes retaining wall displacement, retaining wall stress and retaining wall deformation in the horizontal direction.
[0010] A further improvement of the method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall according to the present invention is that the preprocessing of the monitoring data specifically includes the following steps:
[0011] S11. Process missing values of monitoring data;
[0012] S12. Perform noise reduction on the data after missing value processing.
[0013] A further improvement of the method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall according to the present invention is that step S3 specifically includes the following steps:
[0014] S31. Determine the tuning range of the four hyperparameters: learning rate, filter, convolution kernel size, and pooling size.
[0015] S32. Use the grid search algorithm to traverse all possible combinations of hyperparameters according to the tuning range, and then determine the optimal hyperparameter combination based on the set indicators.
[0016] A further improvement of the method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall of the present invention is that the structure of the model includes a convolution-pooling layer, an LSTM layer, an attention mechanism layer, and a fully connected layer; the output of the convolution-pooling layer serves as the input of the LSTM layer, the output of the LSTM layer serves as the input of the attention mechanism layer, and the output of the attention mechanism layer serves as the input of the fully connected layer;
[0017] The attention mechanism uses dot product to calculate the attention score of each input, which is then converted into attention weight by SoftMax normalization, and then the attention value is obtained by weighted average calculation as the input of the fully connected layer.
[0018] A further improvement of the method for predicting the three-dimensional spatiotemporal distribution of deformation of foundation pit retaining walls of the present invention is that the number of the LSTM layers is multiple, and the output of the convolution-pooling layer is used as the input of the first LSTM layer, and the output of the last LSTM layer is used as the input of the attention mechanism layer.
[0019] Compared with the prior art, the present invention has the following beneficial effects:
[0020] The present invention combines CNN and LSTM to accurately capture spatiotemporal features, and the attention mechanism focuses on key information to improve prediction accuracy. In terms of model optimization, the grid search algorithm efficiently tunes hyperparameters to avoid blind search, improves generalization ability, optimizes resource utilization, reduces costs, improves efficiency, enhances practicality and operability, and provides strong support for retaining wall deformation prediction and safety management in foundation pit projects. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 The flowchart of foundation pit retaining wall deformation prediction of the present invention is shown.
[0022] Figure 2 A schematic diagram of the prediction model structure of the present invention is shown. DETAILED DESCRIPTION
[0023] In order to solve the problem of being unable to predict the deformation of foundation pit retaining walls, the present invention provides a method for predicting the three-dimensional spatiotemporal distribution of foundation pit retaining wall deformation. The following is a further description of this method using specific embodiments and accompanying drawings.
[0024] See Figures 1 and 2 As shown, a method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall comprises the following steps:
[0025] S1. Collect foundation pit monitoring data and pre-process the monitoring data to obtain a spatiotemporal matrix of retaining wall deformation;
[0026] S2. Construct a CNN-LSTM-Att model and input the spatiotemporal matrix into the model. Use a convolutional neural network to extract the spatial features of the retaining wall deformation, use a long short-term memory network to capture the temporal features of the retaining wall deformation, and use the attention mechanism to calculate the attention score of the wall deformation at each time step based on the spatial and temporal features.
[0027] S3. Optimize the model’s hyperparameters;
[0028] S4. Based on the optimized model, the deformation of the retaining wall in the future time step is predicted according to the change pattern of the attention score of the wall deformation at each time step.
[0029] Specifically, CNN stands for convolutional neural network, LSTM stands for long short-term memory network, and Att stands for attention mechanism;
[0030] By adopting the above design, the proposed CNN-LSTM-Att model simultaneously extracts temporal and spatial features from the spatiotemporal feature matrix, achieving superior prediction accuracy. By introducing an attention mechanism, foundation pit monitoring data can be intelligently screened and weighted. For data subject to noise and other interference, the attention mechanism can reduce the impact of unimportant or erroneous data and focus on key, valid data, thereby enhancing the positive impact of data quality on the model. Even with relatively limited data, key features can be learned based on data importance, effectively addressing the constraints of insufficient data quality and quantity on model predictions. This enhances the model's adaptability to diverse data conditions, enabling it to better handle non-uniform data and discover patterns in complex spatiotemporal data. The attention mechanism empowers the model with more powerful processing capabilities for complex spatiotemporal data, enabling it to better capture the key characteristics and changing trends of retaining wall deformation under complex working conditions such as changing geological conditions and construction process interference, thereby improving the model's adaptability and prediction accuracy under various working conditions. At the same time, through the analysis of attention weights, the degree of attention the model pays to different data features during the prediction process can be intuitively presented, overcoming the poor interpretability of the traditional neural network "black box" model, allowing engineers to understand the basis of model decisions, facilitating targeted adjustments and optimizations when complex working conditions change, and improving the applicability of the model.
[0031] Among them, the foundation pit monitoring data includes retaining wall displacement, retaining wall stress and retaining wall deformation in the horizontal direction.
[0032] By obtaining data such as retaining wall displacement, retaining wall stress, and horizontal deformation of the retaining wall, the CNN-LSTM-Att model can be used to accurately predict the three-dimensional spatiotemporal distribution characteristics of retaining wall deformation, providing more accurate information to assist engineering decision-making and risk prevention and control.
[0033] The pre-processing of the monitoring data specifically includes the following steps:
[0034] S11. Process missing values of monitoring data;
[0035] S12. Perform noise reduction on the data after missing value processing.
[0036] By adopting the above design, data integrity and information loss are guaranteed, and the accuracy and reliability of the model are improved.
[0037] Wherein, step S3 specifically includes the following steps:
[0038] S31. Determine the tuning range of the four hyperparameters: learning rate, filter, convolution kernel size, and pooling size.
[0039] S32. Use the grid search algorithm to traverse all possible combinations of hyperparameters according to the tuning range, and then determine the optimal hyperparameter combination based on the set indicators.
[0040] By adopting the above design, the grid search algorithm is used to optimize key hyperparameters such as learning rate, filter, convolution kernel size and pooling size, determine the tuning range, comprehensively search combinations and select the best hyperparameter combination based on the set indicators to ensure the optimal model performance and achieve the effect of optimizing the model.
[0041] Among them, see Figure 2 As shown in the figure, the model structure of the prediction model includes a convolution-pooling layer, an LSTM layer, an attention mechanism layer, and a fully connected layer; the output of the convolution-pooling layer serves as the input of the LSTM layer, the output of the LSTM layer serves as the input of the attention mechanism layer, and the output of the attention mechanism layer serves as the input of the fully connected layer;
[0042] The attention mechanism uses dot product to calculate the attention score of each input, which is then converted into attention weight by normalization through the SoftMax function, and then the attention value is obtained by weighted average calculation as the input of the fully connected layer.
[0043] By clarifying the functional division of each component in the model, such as the convolution-pooling layer is responsible for extracting spatial features, the LSTM layer is responsible for capturing temporal features, and the attention mechanism enhances the focus and processing capabilities of key information, and each part works together to achieve highly stable performance; highlighting key features reduces prediction errors, allowing the model to selectively highlight key information from different inputs and adapt to long and short-term input sequences.
[0044] Among them, see Figure 2 As shown in the figure, there are multiple LSTM layers, and the output of the convolution-pooling layer is used as the input of the first LSTM layer, and the output of the last LSTM layer is used as the input of the attention mechanism layer;
[0045] Furthermore, in this embodiment, the number of LSTM layers does not exceed 8.
[0046] By adopting the above design, the model can be trained at a deeper level.
[0047] Based on the above-mentioned three-dimensional spatiotemporal distribution prediction method of foundation pit retaining wall deformation, a foundation pit retaining wall deformation risk assessment method is proposed, which includes the following steps:
[0048] Assuming that the risk associated with retaining wall deformation follows a normal distribution, the risk is divided into multiple risk levels using equal percentage intervals;
[0049] Obtain the maximum deformation rate index of the retaining wall in the foundation pit project over the past week, and determine the risk level corresponding to this index by referring to equal percentage intervals of multiple risk levels;
[0050] The risk level is determined by obtaining the median standard deviation of the attention weight of the retaining wall in the foundation pit project in the past week. The standard deviation of this indicator reflects the fluctuation of the deflection rate of the retaining wall in the past week.
[0051] Through the model's in-depth understanding and processing of data, these indicators accurately reflect the risk of retaining wall deformation and effectively guide on-site construction safety management. Based on this indicator, engineers can promptly adjust construction strategies and implement protective measures to ensure the safe and smooth progress of foundation pit projects. This addresses the shortcomings of previous technologies in construction safety management, allowing the model to move beyond prediction and become deeply integrated into the entire engineering practice process. Using the median to mitigate the influence of extreme values, the combination of the two indicators improves risk assessment accuracy and reduces model prediction errors and instability.
[0052] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.
[0053] The present invention has been described in detail above with reference to the embodiments of the accompanying drawings. A person skilled in the art can make various modifications to the present invention based on the above description. Therefore, certain details in the embodiments should not be construed as limiting the present invention. The scope of protection of the present invention shall be determined by the scope defined in the appended claims.
Claims
1. A method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall, characterized in that: The steps include: S1. Collect foundation pit monitoring data and pre-process the monitoring data to obtain a spatiotemporal matrix of retaining wall deformation; S2. Construct a CNN-LSTM-Att model, input the spatiotemporal matrix into the model, use a convolutional neural network to extract the spatial features of the retaining wall deformation, use a long short-term memory network to capture the temporal features of the retaining wall deformation, and use an attention mechanism to calculate the attention score of the wall deformation at each time step based on the spatial features and the temporal features; S3. Optimize the model’s hyperparameters; S4. Based on the optimized model, the deformation of the retaining wall in the future time step is predicted according to the change pattern of the attention score of the wall deformation at each time step.
2. The method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall according to claim 1, wherein: The foundation pit monitoring data includes retaining wall displacement, retaining wall stress, and retaining wall deformation in the horizontal direction.
3. The method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall according to claim 3, wherein: The pre-processing of the monitoring data specifically comprises the following steps: S11. Process missing values of monitoring data; S12. Perform noise reduction on the data after missing value processing.
4. The method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall according to claim 1, wherein: The step S3 specifically includes the following steps: S31. Determine the tuning range of the four hyperparameters: learning rate, filter, convolution kernel size, and pooling size. S32. Use the grid search algorithm to traverse all possible combinations of hyperparameters according to the tuning range, and then determine the optimal hyperparameter combination based on the set indicators.
5. The method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall according to claim 1, wherein: The structure of the model includes a convolution-pooling layer, an LSTM layer, an attention mechanism layer, and a fully connected layer; the output of the convolution-pooling layer is used as the input of the LSTM layer, the output of the LSTM layer is used as the input of the attention mechanism layer, and the output of the attention mechanism layer is used as the input of the fully connected layer; The attention mechanism uses dot product to calculate the attention score of each input, which is then converted into attention weight by SoftMax normalization, and then the attention value is obtained by weighted average calculation as the input of the fully connected layer.
6. The method for predicting the three-dimensional spatiotemporal distribution of deformation of a foundation pit retaining wall according to claim 5, characterized in that: There are multiple LSTM layers, and the output of the convolution-pooling layer is used as the input of the first LSTM layer, and the output of the last LSTM layer is used as the input of the attention mechanism layer.