A foundation pit deformation prediction method, system, electronic device and storage medium
By constructing an AM-CNN-LSTM model and training a foundation pit deformation prediction model using multi-source monitoring data, the problems of high computational cost and insufficient spatial feature modeling in existing technologies are solved, and high-precision prediction of foundation pit deformation and identification of safety risks are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA HYDROPOWER CONSTR GRP INT ENG CO LTD
- Filing Date
- 2026-04-07
- Publication Date
- 2026-06-12
Smart Images
Figure CN122196549A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of foundation pit deformation prediction technology, and in particular to a foundation pit deformation prediction method, system, electronic device and storage medium. Background Technology
[0002] Foundation pit engineering plays a crucial role in the development of urban underground space. Its deformation evolution is a complex, nonlinear, dynamic, spatiotemporal process influenced by a combination of factors, including geological and hydrological conditions, support systems, and excavation construction. Constructing a high-precision deformation prediction model has significant theoretical and engineering value for proactively identifying safety risks and guiding information-based construction.
[0003] Currently, existing technologies for predicting foundation pit deformation mainly focus on numerical simulation. However, numerical simulation-based prediction methods involve large computational loads and cannot effectively model the spatial characteristics of foundation pit deformation, resulting in an inability to accurately predict foundation pit deformation. Summary of the Invention
[0004] In view of this, embodiments of this application provide a method, system, electronic device, and storage medium for predicting foundation pit deformation, which can improve the accuracy of foundation pit deformation prediction.
[0005] In a first aspect, embodiments of this application provide a method for predicting foundation pit deformation. The method includes: acquiring multi-source monitoring data of the foundation pit to be tested and preprocessing it to construct a training dataset; training a pre-constructed foundation pit deformation prediction model based on the training dataset, wherein the pre-constructed foundation pit deformation prediction model includes an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer; the input layer is used to receive input features from the training dataset; the convolutional neural network module is used to extract local features from the input features and output a feature vector; the bidirectional long short-term memory network is used to receive the feature vector, capture the long-term and short-term dependencies of foundation pit deformation over time, and output a hidden state sequence; the attention module is used to perform adaptive weighted calculation on the hidden state sequence to generate a weighted feature vector; the output layer is used to output predicted deformation data generated by flattening the weighted feature vector and performing a nonlinear transformation through a fully connected layer; acquiring collected real-time monitoring data and inputting the real-time monitoring data into the trained pre-constructed foundation pit deformation prediction model to obtain the predicted deformation data of the foundation pit to be tested.
[0006] Optionally, the step of acquiring historical monitoring data of the foundation pit to be tested and preprocessing it to construct a training dataset includes: acquiring multi-source monitoring data of the foundation pit site to be tested; analyzing the multi-source monitoring data based on grey relational analysis to obtain key feature data; the multi-source monitoring data includes historical deformation state variables, environmental factors, geological parameters, and construction factors; performing polynomial interpolation on the key feature data to obtain processed key feature data; normalizing the processed key feature data to construct an original dataset; and using the sliding window method to reconstruct samples from the original dataset, using monitoring data from consecutive historical moments as input features and deformation data from the next moment as prediction labels to form several training samples and construct a training dataset.
[0007] Optionally, training the pre-built foundation pit deformation prediction model based on the training dataset includes: inputting training samples from the training dataset into the pre-built foundation pit deformation prediction model; extracting local features from the input features using the convolutional neural network module to obtain a feature vector; performing temporal evolution on the feature vector using the bidirectional long short-term memory network module to output a hidden state sequence; performing adaptive weighted calculation on the hidden state sequence using the attention module to generate a weighted vector; performing forward propagation on the weighted vector to calculate the error between the predicted foundation pit deformation value and the actual value; and iteratively updating the model parameters using a backpropagation algorithm and an adaptive distance estimation optimization algorithm until a preset convergence condition is met to obtain the trained pre-built foundation pit deformation prediction model.
[0008] Optionally, the historical deformation state variables include at least the axial force of the supports and the horizontal displacement of the wall; the environmental factors include at least the groundwater volume, rainfall, and temperature; the geological parameters include at least the cohesion, internal friction angle, horizontal permeability coefficient, and average compression modulus of the soil layer; and the construction factors include at least the excavation depth of the foundation pit and the number of cross braces.
[0009] Optionally, the method further includes: comparing the predicted deformation data with a preset warning threshold; if the predicted deformation data exceeds the preset warning threshold, triggering a corresponding warning signal.
[0010] Secondly, embodiments of this application also provide a foundation pit deformation prediction system, the system comprising: a dataset construction module, used to acquire multi-source monitoring data of the foundation pit to be tested, and preprocess it to construct a training dataset; a model training module, used to train a pre-constructed foundation pit deformation prediction model based on the training dataset, wherein the pre-constructed foundation pit deformation prediction model includes an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer; the input layer is used to receive input features from the training dataset; the convolutional neural network module is used to perform local feature extraction on the input features, and output... The system generates a feature vector; the bidirectional long short-term memory network receives the feature vector, captures the long- and short-term dependencies of the foundation pit deformation over time, and outputs a hidden state sequence; the attention module performs adaptive weighted calculation on the hidden state sequence to generate a weighted feature vector; the output layer outputs the flattened weighted feature vector, which is then subjected to a nonlinear transformation by a fully connected layer to generate predicted deformation data; and the foundation pit deformation prediction module acquires real-time monitoring data and inputs the real-time monitoring data into the pre-built foundation pit deformation prediction model to obtain the predicted deformation data of the foundation pit to be tested.
[0011] Optionally, the dataset construction module includes: a data acquisition module, used to acquire multi-source monitoring data from the site of the foundation pit to be tested, and analyze the multi-source monitoring data based on grey relational analysis to obtain key feature data; the multi-source monitoring data includes historical deformation state variables, environmental factors, geological parameters, and construction factors; a first processing submodule, used to perform polynomial interpolation processing on the key feature data to obtain processed key feature data; a second processing submodule, used to normalize the processed key feature data to construct the original dataset; and a dataset construction submodule, used to reconstruct samples from the original dataset using the sliding window method, taking monitoring data from continuous historical moments as input features and deformation at the next moment as prediction labels to form several training samples and construct a training dataset.
[0012] Optionally, the model training module is specifically used to input training samples from the training dataset into the pre-built foundation pit deformation prediction model; extract local features from the input features based on the convolutional neural network module to obtain a feature vector; perform temporal evolution on the feature vector based on the bidirectional long short-term memory network module to output a hidden state sequence; perform adaptive weighted calculation on the hidden state sequence based on the attention module to generate a weighted vector; perform forward propagation on the weighted vector to calculate the error between the predicted foundation pit deformation value and the true value; and iteratively update the model parameters through backpropagation algorithm and adaptive distance estimation optimization algorithm until a preset convergence condition is met to obtain the trained pre-built foundation pit deformation prediction model.
[0013] Optionally, the historical deformation state variables include at least the axial force of the supports and the horizontal displacement of the wall; the environmental factors include at least the groundwater volume, rainfall, and temperature; the geological parameters include at least the cohesion, internal friction angle, horizontal permeability coefficient, and average compression modulus of the soil layer; and the construction factors include at least the excavation depth of the foundation pit and the number of cross braces.
[0014] Optionally, the system further includes an early warning module, used to compare the predicted deformation data with a preset early warning threshold, and if the predicted deformation data exceeds the preset early warning threshold, trigger a corresponding early warning signal.
[0015] Thirdly, this application also provides an electronic device, which includes: a housing, a processor, a memory, a circuit board, and a power supply circuit, wherein the circuit board is disposed inside the space enclosed by the housing, and the processor and the memory are disposed on the circuit board; the power supply circuit is used to supply power to various circuits or devices of the above-mentioned electronic device; the memory is used to store executable program code; the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, for executing the foundation pit deformation prediction method described in any of the first aspects above.
[0016] Fourthly, embodiments of this application provide a computer-readable storage medium storing one or more programs, which can be executed by one or more processors to implement the foundation pit deformation prediction method described in any of the first aspects.
[0017] This application provides a method, system, electronic device, and storage medium for predicting foundation pit deformation. It constructs a foundation pit deformation prediction model including an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer. The convolutional neural network module extracts local features from the input monitoring data and outputs a feature vector. The bidirectional long short-term memory network captures the long-term and short-term dependencies of foundation pit deformation over time, outputting a hidden state sequence. Finally, the attention module performs adaptive weighted calculations on the hidden state sequence to generate a weighted feature vector. The weighted feature vector is then decoded to output the generated predicted deformation data, significantly improving the accuracy of foundation pit deformation prediction. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a schematic diagram of a foundation pit deformation prediction method provided in an embodiment of this application;
[0020] Figure 2 A flowchart illustrating step S110 of the foundation pit deformation prediction method provided in an embodiment of this application;
[0021] Figure 3 A flowchart illustrating step S120 of the foundation pit deformation prediction method provided in an embodiment of this application;
[0022] Figure 4 This is a schematic diagram of the foundation pit deformation prediction model architecture provided in an embodiment of this application;
[0023] Figure 5 This is a schematic diagram of a foundation pit monitoring point provided in one embodiment of this application;
[0024] Figure 6 This is a schematic diagram of the foundation pit deformation prediction system architecture provided in one embodiment of this application;
[0025] Figure 7 This is a schematic block diagram illustrating the architecture of an embodiment of the electronic device of this application. Detailed Implementation
[0026] The embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0027] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.
[0028] See Figure 1 This application provides a method for predicting foundation pit deformation, including the following steps:
[0029] S110. Obtain multi-source monitoring data of the foundation pit to be tested, and preprocess it to construct a training dataset.
[0030] In this step, based on the automated detection system of the foundation pit to be tested, or sensors deployed at different monitoring points on the foundation pit, multi-source monitoring data of the foundation pit from the start of excavation to the current stage are collected. The collection frequency can be once a day. The collected multi-source monitoring data is preprocessed to construct the dataset for model training. During the construction of the foundation pit, the deformation of the foundation pit is comprehensively affected by geological conditions, construction factors, and environmental factors. Therefore, the collected multi-source monitoring data includes historical deformation state variables, environmental factors, geological parameters, and construction factors. By collecting multi-source monitoring data during the trial process, the comprehensiveness and timeliness of the data source are ensured, providing an efficient and high-quality data source for subsequent data processing and model construction.
[0031] In some embodiments, see Figure 2 Step S110: Obtain multi-source monitoring data of the foundation pit to be tested, and preprocess it to construct a training dataset, including:
[0032] S111. Obtain multi-source monitoring data of the foundation pit to be tested, and analyze the multi-source monitoring data based on grey relational analysis to obtain key feature data; the multi-source monitoring data includes historical deformation state variables, environmental factors, geological parameters, and construction factors; wherein, the historical deformation state variables include at least the axial force of the supports and the horizontal displacement of the walls; the environmental factors include at least the groundwater volume, rainfall, and temperature; the geological parameters include at least the cohesion, internal friction angle, horizontal permeability coefficient, and average compression modulus of the soil layer; the construction factors include at least the excavation depth of the foundation pit and the number of cross braces;
[0033] S112. Perform polynomial interpolation on the key feature data to obtain the processed key feature data;
[0034] S113. Normalize the processed key feature data to construct the original dataset;
[0035] S114. The original dataset is reconstructed using the sliding window method. The monitoring data of continuous historical moments are used as input features, and the deformation data of the next moment is used as the prediction label to form several training samples and construct a training dataset.
[0036] In this embodiment, after obtaining multi-source monitoring data from the on-site monitoring points of the foundation pit to be tested, the correlation of the multi-source monitoring data obtained from the on-site monitoring points of the foundation pit to be tested is first calculated using the grey relational analysis method. Features with low correlation are eliminated, and key feature data is retained. The key feature data in this application includes: historical maximum deformation state variables containing the axial force of the supports and the deep horizontal displacement of the wall; geological factors containing cohesion, internal friction angle, horizontal permeability coefficient and average compression modulus of the soil layer; construction factors containing the excavation depth of the foundation pit and the number of cross braces; and environmental factors. Then, the key feature data is preprocessed, including missing value processing, outlier detection and normalization. Since some data may be missing in the original key feature data, this... The application embodiment employs polynomial interpolation to perform sequence transformation on the original key feature data, filling in missing values to ensure data integrity and continuity. Furthermore, since the original key feature data may contain some unrealistic situations, this application uses the 3σ criterion for outlier detection and removal to ensure data accuracy, generating a complete continuous time series containing all features of the key feature data, i.e., the processed key feature data. Subsequently, since the monitoring data from different monitoring points differ significantly in value, this application embodiment uses the min-max normalization method to normalize the data, ensuring data consistency, constructing the original dataset, and dividing the data in the original dataset into a training set, a validation set, and a test set according to time in a 5:2:3 ratio.
[0037] Next, a sliding window method is used to construct a training sample dataset. The data in the normalized original training set is split into input features and output labels to reconstruct the sample format, forming several training samples and constructing the training dataset. The input features include monitoring data from continuous historical moments, for example, using the key features of the first 5 consecutive time steps as input features. The output labels include the deformation data of the next moment, using the target deformation state variable of the 6th time step as the output prediction label. That is, the key monitoring data of the key monitoring point in the previous 5 days is used as input data, and the monitoring data of the monitoring point on the 6th day is predicted as output prediction data.
[0038] S120. Train a pre-built foundation pit deformation prediction model based on the training dataset, wherein the pre-built foundation pit deformation prediction model includes a convolutional neural network module, a bidirectional long short-term memory network module, and an attention module.
[0039] In this step, a pre-built foundation pit deformation prediction model is trained using training samples from the constructed training sample dataset. The predicted deformation data of the foundation pit is compared with the measured deformation data of the foundation pit. The parameters of the foundation pit deformation prediction model are adjusted according to the prediction error. The model training process is repeated until the test set loss converges and no longer decreases or reaches the set number of iterations. Training is then stopped, and the pre-built foundation pit deformation prediction model is output. The pre-built foundation pit deformation prediction model is a prediction model based on attention mechanism-convolutional neural network-bidirectional long short-term memory network (AM-CNN-LSTM), including an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer.
[0040] See Figure 4 The pre-built foundation pit deformation prediction model is based on AM-CNN-LSTM, comprising an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer. The input layer receives input features from preprocessed training samples, i.e., a multi-dimensional input feature matrix including deformation state variables, geological parameters, construction factors, and environmental factors. The core structure includes a convolutional neural network module, a bidirectional long short-term memory network module, and an attention module. The convolutional neural network module includes a one-dimensional convolutional layer and a pooling layer. The one-dimensional convolutional layer extracts local features from the multi-dimensional input features, and the pooling layer downsamples the time steps, preserving key features and reducing computational cost to finally output a feature vector. The bidirectional long short-term memory network module includes three LSTM units to receive feature vectors and capture the long-term and short-term dependencies (Ci, Ci) of foundation pit deformation over time. t / C t-1 ), output the hidden state sequence (h t / h t-1 / h t+1 The attention module employs a multiplicative attention mechanism to adaptively weight the hidden state sequence (w in the diagram). t w gt The attention mechanism uses learnable weight parameters to adjust the contribution of different input features, generating a weighted feature vector (context vector). The output layer flattens the weighted feature vector output by the attention module, performs a non-linear transformation through a fully connected layer, and finally generates the predicted deformed data (y) for the next time step. t+1 ).
[0041] In this embodiment, the one-dimensional convolutional layer in the convolutional neural network module has a kernel size of 3, a channel count of 64, a stride of 1, and uses the Rectified Linear Activation Function (ReLU) to filter out high-frequency random noise and enhance nonlinear expressive power. The pooling layer has a pooling window of 2 and a stride of 1, and uses max pooling to compress the data dimension while retaining significant feature responses. The bidirectional long short-term memory network module has a single-layer LSTM (Long Short-Term Memory) structure, including three LSTM units, and 64 hidden layer neurons. The attention module has a multiplicative attention mapping dimension of 20 and initializes a learnable weight matrix.
[0042] In some embodiments, see Figure 3 Step S120: Train a pre-built foundation pit deformation prediction model based on the training dataset, including:
[0043] S121. Input the training samples in the training dataset into the pre-built foundation pit deformation prediction model, and extract local features from the key feature data based on the convolutional neural network module to obtain feature vectors;
[0044] S122. Based on the bidirectional long short-term memory network module, the feature vector is subjected to temporal evolution to output the hidden state sequence;
[0045] S123. Based on the attention module, perform adaptive weighted calculation on the hidden state sequence to generate a weighted vector;
[0046] S124. Perform forward propagation on the weighted vector to calculate the error between the predicted value and the actual value of the foundation pit deformation. Then, iteratively update the model parameters through the backpropagation algorithm and the adaptive distance estimation optimization algorithm until the preset convergence condition is met, and obtain the trained pre-constructed foundation pit deformation prediction model.
[0047] In this embodiment, during model training, the mean squared error (MSE) is selected as the loss function, and the Adam optimizer is used to perform global parameter optimization. The initial learning rate is set to 0.001, and the maximum number of training iterations is set to 100. Data from the training dataset is input into the model, and the batch size and number of iterations are set. The predicted deformation data of the foundation pit are compared with the measured deformation data of the foundation pit. The parameters of the foundation pit deformation prediction model are adjusted according to the prediction error. The model training process is repeated until the loss on the test set converges and no longer decreases or the set number of iterations is reached. Training is then stopped, and the pre-built foundation pit prediction model is output.
[0048] Specifically, the key feature data from the training dataset is input into the convolutional neural network module through the input layer. A one-dimensional convolutional kernel is used to extract local features from the key feature data, outputting a high-dimensional feature vector. This high-dimensional feature vector is then expanded over time steps and input into a bidirectional long short-term memory (LSTM) network module to learn the temporal evolution pattern. The cell state and hidden state at each time step are calculated and transmitted. The information flow is finely regulated through the control coefficients of the forget gate, input gate, and output gate, outputting a sequence of hidden states fused with LTM. This hidden state sequence is then passed to the attention mechanism layer, where the correlation score between the current predicted deformation data and the deformation state variable is calculated to update the global weights. Normalized weight coefficients are generated using the Softmax function, constructing a weighted vector containing global spatiotemporal information, also known as a context vector. Finally, the weighted vector is used to calculate the error between the predicted and actual values of the foundation pit deformation through the fully connected layer of the output layer. The model parameters are then iteratively updated using backpropagation and adaptive distance estimation optimization algorithms until the preset convergence condition is met, resulting in a pre-trained, pre-constructed foundation pit deformation prediction model. The preset convergence conditions are reaching the maximum cycle length of 100 times or the validation set loss remaining stable without decreasing.
[0049] S130. Acquire the collected real-time monitoring data and input the real-time monitoring data into the pre-built foundation pit deformation prediction model to obtain the predicted deformation data of the foundation pit to be tested.
[0050] In this embodiment, after the model training is completed, the optimal model weights are saved. During the actual prediction stage, real-time monitoring data from monitoring points is collected, normalized, and then input into the pre-trained, pre-built foundation pit deformation prediction model. Finally, the normalized prediction data is output, and then linearly inversely mapped back to the actual monitored physical quantity size.
[0051] In some embodiments, the method further includes: analyzing the predicted deformation data of the foundation pit to be tested. Specifically, the foundation pit deformation prediction model provided in this embodiment is compared and verified with a single CNN model, a single LSTM model, and a CNN-LSTM benchmark model. The mean absolute error (MAE) and root mean square error (RMSE) are used to comprehensively and quantitatively judge the overall fitting performance of the foundation pit deformation prediction model in this embodiment. The results show that the mean absolute error and root mean square error of the foundation pit deformation prediction model based on AM-CNN-LSTM in this embodiment are significantly reduced, improving the accuracy of foundation pit deformation prediction.
[0052] In some embodiments, the method further includes: comparing the predicted deformation data with a preset warning threshold; if the predicted deformation data exceeds the preset warning threshold, triggering a corresponding warning signal.
[0053] In this embodiment, the predicted deformation data of the foundation pit output by the model is compared with the preset early warning threshold of the actual project. If the predicted deformation data exceeds the preset early warning threshold, a corresponding early warning signal is triggered. For example, assuming the alarm value of the horizontal displacement of the top of the foundation pit wall is 30mm, when the model predicts that the deformation value will reach 24mm (80% of the alarm value) in the next day, the system triggers a yellow warning; when the predicted value reaches 27mm (90% of the alarm value), an orange warning is triggered, prompting the on-site engineer to intensify monitoring, analyze the cause, and prepare emergency measures.
[0054] The foundation pit deformation prediction method provided in this embodiment predicts foundation pit deformation by constructing a foundation pit deformation prediction model including an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer. The convolutional neural network module extracts local features from the input monitoring data and outputs a feature vector. The bidirectional long short-term memory network captures the long-term and short-term dependencies of foundation pit deformation over time and outputs a hidden state sequence. Finally, the attention module performs adaptive weighted calculation on the hidden state sequence to generate a weighted feature vector. The weighted feature vector is then decoded to output the predicted deformation data, significantly improving the accuracy of foundation pit deformation prediction.
[0055] The following specific embodiment illustrates the foundation pit deformation prediction method provided in this application:
[0056] See Figure 5 In this embodiment, a prediction experiment was conducted at monitoring points 35, 55, and 65 in a network dispatch and command building, including the following steps:
[0057] S1. Constructing a dataset: Obtain historical multi-source heterogeneous monitoring data from monitoring points, and preprocess the data, perform feature filtering and kernel partitioning to construct a dataset; the multi-source heterogeneous monitoring data includes environmental factors, geological parameters, construction factors, and deformation state variables.
[0058] S11. By querying data sources and synchronously recording data of multiple indicators at the foundation pit site, a time-series dataset is formed, in which... For a point in time, For the corresponding monitoring data, polynomial interpolation is used to transform the original time series data, fill in missing values, and generate a continuous time series. The steps of the polynomial interpolation method include:
[0059] We construct higher-order Lagrange interpolation polynomial basis functions to capture the nonlinear evolution of the data, expressed as:
[0060] In the formula, Let j be the Lagrange interpolation basis function, and j be the loop variable, iterating through all node indices except the current k. This represents the time coordinate of the j-th node; This represents the time coordinate of the selected interpolation node, where t is the time of interpolation.
[0061] For the time t to be interpolated, substitute it into the interpolation polynomial to calculate the corresponding interpolation, and obtain the continuous time series, expressed by the formula:
[0062] In the formula, represents the interpolation polynomial; t represents the time point to be interpolated. This represents the known monitoring data at the k-th node. Let the Lagrange interpolation basis function satisfy the following condition for the k-th node: The value is 1 at time nodes and 0 at other nodes.
[0063] S12. Construct a multi-dimensional precursor feature system, introduce grey relational analysis to calculate the correlation of multi-source heterogeneous monitoring data including environmental factors, geological parameters, construction factors and deformation state variables, perform dimensionality reduction screening on all features, remove variables with correlation degree less than 0.5, and finally retain 8 core feature data: cohesion, internal friction angle, horizontal permeability coefficient, average compression modulus of soil layer, excavation depth, number of cross braces, support axial force and deep horizontal displacement of wall.
[0064] S13. Divide the expanded data into training, validation, and test sets according to time in a 5:2:3 ratio. Normalize the training set data. To eliminate the influence of different units, use the Min-MaxNormalization method to scale all feature data to the [0, 1] interval. The calculation formula is as follows:
[0065] In the formula, x is the original value. The minimum value of this feature across the entire training set. This is the maximum value of the feature across the entire training set.
[0066] S14. The normalized data is split into multidimensional input tensors and output labels. The sample format is reconstructed using the sliding window technique. The input tensor contains 8 feature data items from the first 5 time steps, and the output is the target deformation state variable from the 6th time step that follows.
[0067] S2. Construct a foundation pit deformation prediction model based on AM-CNN-LSTM: Construct a foundation pit deformation prediction model that includes an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer.
[0068] Specifically, the input layer receives input features from preprocessed training samples, including multidimensional input features such as deformation state variables, geological parameters, and construction factors. The core structure includes a convolutional neural network module, a bidirectional long short-term memory (LSTM) network module, and an attention module. The convolutional neural network module comprises a one-dimensional convolutional layer and a pooling layer. The one-dimensional convolutional layer extracts local features from the multidimensional input features, while the pooling layer downsamples the time steps, preserving key features and reducing computational complexity to output a feature vector. The bidirectional LSTM network module includes three LSTM units to receive feature vectors, capture the long-term and short-term dependencies of the foundation pit deformation over time, and output a hidden state sequence. The attention module employs a multiplicative attention mechanism to adaptively weight the hidden state sequence, generating a weighted feature vector. The output layer flattens the weighted feature vector output by the attention module and performs a nonlinear transformation through a fully connected layer to generate the predicted deformation data for the next time step. In this embodiment, the dimension of a single input sample tensor is (5,8), representing 5 time steps and 8 feature indicators.
[0069] S3. Initialize the model;
[0070] S31. In the convolutional neural network module, the kernel size of the one-dimensional convolutional layer is 3, the number of channels is 64, the stride is set to 1, and the activation function is the rectified linear function (ReLU) to filter out high-frequency random noise and enhance nonlinear expression ability; the pooling window in the pooling layer is set to 2, the stride is set to 1, and the max pooling layer is used to compress the data dimension and retain significant feature responses.
[0071] S32. Set up a single-layer LSTM structure in the bidirectional long short-term memory network module, including three LSTM units, with 64 hidden layer neurons.
[0072] S33. In the attention module, set the multiplicative attention mapping dimension to 20 and initialize the learnable weight matrix. Use the Adam optimizer to perform global parameter optimization, setting the initial learning rate to 0.001 and the maximum number of training iterations to 100.
[0073] S4. Model Training:
[0074] S41. The input layer inputs the reconstructed multidimensional feature tensor into the preceding convolutional neural network module, using one-dimensional convolution and pooling operations to filter out high-frequency random noise and extract the hidden spatial feature vectors; where the convolution calculation formula is expressed as: y=f(W x+b), where, denoted by , where W is the convolution kernel weight matrix, b is the bias term, and f is the activation function. In this embodiment, the rectified linear function (ReLU) is used.
[0075] S42. Expand the spatial feature vector output by the convolutional neural network module by time step and input it into the bidirectional long short-term memory network module. The gated recurrent unit captures the long-term dependence and cumulative effect of deformation evolution, calculates and outputs the hidden state sequence. ;
[0076] S43, Attention module receives hidden state sequence The multiplicative attention mechanism is used to calculate the correlation score between the current prediction task and the historical deformed state variables, and then the global weights are updated. The formula for calculating the correlation score is as follows: In the formula, h is a learnable weight matrix. t h is the current hidden state vector. i e is the historical hidden state vector; t,i A correlation score is assigned to measure spatiotemporal correlation.
[0077] Normalized weight coefficients are generated using the Softmax function. Construct a context vector containing global spatiotemporal information: In the formula, The normalized attention weights represent the contribution of historical features; c t For the context vector, i.e., the weighted and aggregated global features, h i This is the historical hidden state vector.
[0078] S44. The context vector is used to calculate the error distribution between the predicted and true values through the fully connected layer of the output layer. The Adam optimization algorithm is used to perform backpropagation of network parameters. During the training process, the validation set data is used periodically to evaluate the generalization performance of the model to prevent overfitting.
[0079] S45. Repeat the supervised training operation on the entire dataset according to the above steps S41-S44 until the maximum number of iterations of 100 is reached or the loss on the validation set stabilizes and does not decrease, then terminate the training process.
[0080] S5. Prediction of foundation pit deformation:
[0081] S51. Input the collected real-time monitoring data into the prepared foundation pit deformation prediction model. Output the final normalized foundation pit deformation state variable prediction data through the output layer. Perform a linear inverse mapping on the normalized output values to restore them to the actual monitored physical quantity dimensions. The formula is:
[0082] In the formula, x is the actual monitored value. This is the data for predicting the deformation state variables of the foundation pit. The minimum value of this feature across the entire training set. This is the maximum value of the feature across the entire training set.
[0083] S52. Based on the actual working conditions of the project and the test data, set the safety early warning threshold for the deformation state variables of the foundation pit at the corresponding monitoring points;
[0084] S53. By combining advanced prediction values, dynamic feedback of construction hazards and linkage of safety early warning are realized on the monitoring platform.
[0085] The foundation pit deformation prediction method provided in this application is based on AM-CNN-LSTM to construct a foundation pit deformation prediction model. The convolutional neural network module, with its local perception mechanism, accurately analyzes the spatial distribution pattern in the monitoring data, extracts and reduces the spatial coupling features of foundation pit deformation. The long short-term memory network module receives the spatial feature sequence output by the front layer, processes the time series information through a gated recurrent structure, and effectively captures the dynamic law and long-term dependency relationship of foundation pit deformation as construction progresses. The attention module addresses the information dilution bottleneck in long sequence transmission by dynamically calculating and allocating attention weights, adaptively focusing on key moments that induce abrupt changes in state variables, such as excavation and unloading, which greatly improves the model's feature capture capability under non-stationary working conditions.
[0086] Meanwhile, the foundation pit deformation prediction method of this application avoids the limitations of a single model in feature extraction. It reduces and lowers the dimensionality of input features by using polynomial interpolation and grey relational analysis, effectively breaking through the bottleneck of traditional methods in processing high-dimensional nonlinear monitoring data. It significantly improves the robustness and generalization accuracy of prediction, and can realize safe dynamic monitoring and early warning of potential dangers in foundation pit engineering.
[0087] See Figure 6 This application also provides a foundation pit deformation prediction system, including:
[0088] The dataset construction module 61 is used to acquire multi-source monitoring data of the foundation pit to be tested and to preprocess the data to construct a training dataset.
[0089] The model training module 62 is used to train a pre-constructed foundation pit deformation prediction model based on the training dataset. The pre-constructed foundation pit deformation prediction model includes an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer. The input layer receives input features from the training dataset. The convolutional neural network module extracts local features from the input features and outputs a feature vector. The bidirectional long short-term memory network receives the feature vector, captures the long-term and short-term dependencies of foundation pit deformation over time, and outputs a hidden state sequence. The attention module performs adaptive weighted calculations on the hidden state sequence to generate a weighted feature vector. The output layer outputs predicted deformation data by flattening the weighted feature vector and then performing a non-linear transformation through a fully connected layer.
[0090] The foundation pit deformation prediction module 63 is used to acquire the collected real-time monitoring data and input the real-time monitoring data into the pre-built foundation pit deformation prediction model to obtain the predicted deformation data of the foundation pit to be tested.
[0091] In some embodiments, the dataset construction module includes: a data acquisition module, used to acquire multi-source monitoring data from the site of the foundation pit to be tested, and to analyze the multi-source monitoring data based on grey relational analysis to obtain key feature data; the multi-source monitoring data includes historical deformation state variables, environmental factors, geological parameters, and construction factors;
[0092] The first processing submodule is used to perform polynomial interpolation on the key feature data to obtain the processed key feature data.
[0093] The second processing submodule is used to normalize the processed key feature data and construct the original dataset.
[0094] The dataset construction submodule is used to reconstruct samples from the original dataset using the sliding window method. It takes the monitoring data from consecutive historical moments as input features and the deformation data from the next moment as prediction labels to form several training samples and construct a training dataset.
[0095] In some embodiments, the model training module is specifically used to input training samples from the training dataset into the pre-built foundation pit deformation prediction model, and to extract local features from the input features based on the convolutional neural network module to obtain a feature vector;
[0096] The feature vector is subjected to temporal evolution based on the bidirectional long short-term memory network module, and the hidden state sequence is output.
[0097] Based on the attention module, an adaptive weighted calculation is performed on the hidden state sequence to generate a weighted vector;
[0098] The weighted vector is forward propagated to calculate the error between the predicted and actual values of the foundation pit deformation. The model parameters are then iteratively updated using a backpropagation algorithm and an adaptive distance estimation optimization algorithm until the preset convergence condition is met, thus obtaining a pre-built foundation pit deformation prediction model that has been trained.
[0099] In some embodiments, the historical deformation state variables include at least the axial force of the supports and the horizontal displacement of the wall; the environmental factors include at least the groundwater volume, rainfall, and temperature; the geological parameters include at least the cohesion, internal friction angle, horizontal permeability coefficient, and average compression modulus of the soil layer; and the construction factors include at least the excavation depth of the foundation pit and the number of cross braces.
[0100] In some embodiments, the system further includes: an early warning module, configured to compare the predicted deformation data with a preset early warning threshold, and if the predicted deformation data exceeds the preset early warning threshold, trigger a corresponding early warning signal.
[0101] The system in this embodiment can be used to execute the technical solution of the foundation pit deformation prediction method shown in the foregoing embodiments. Its implementation principle and technical effect are similar. For details not described in detail, please refer to each other. It will not be repeated here.
[0102] See Figure 7 This application also provides an electronic device. Based on the same technical concept as the foregoing embodiments, the electronic device provided in this application can implement the steps and flow of any of the embodiments described in this application.
[0103] The aforementioned electronic device may include: a housing 71, a processor 72, a memory 73, a circuit board 74, and a power supply circuit 75, wherein the circuit board 74 is disposed inside the space enclosed by the housing 71, and the processor 72 and the memory 73 are disposed on the circuit board 74; the power supply circuit 75 is used to supply power to the various circuits or devices of the aforementioned electronic device; the memory 73 is used to store executable program code; the processor 72 runs the program corresponding to the executable program code by reading the executable program code stored in the memory 73, for executing any of the foundation pit deformation prediction methods described in the aforementioned embodiment one.
[0104] For details on the specific execution process of the above steps by the processor 72 and the steps further executed by the processor 72 by running executable program code, please refer to the description of Embodiment 1 of this application, which will not be repeated here.
[0105] The electronic device exists in various forms, including but not limited to: (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and are primarily designed to provide voice and data communication. These terminals include smartphones (such as iPhones), multimedia phones, feature phones, and low-end phones.
[0106] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, possessing computing and processing capabilities, and generally also have mobile internet access features. These terminals include PDAs, MIDs, and UMPCs, such as the iPad.
[0107] (3) Portable entertainment devices: These devices can display and play multimedia content. This category includes audio and video players (such as iPods), handheld game consoles, e-book readers, as well as smart toys and portable car navigation devices.
[0108] (4) Server: A device that provides computing services. The components of a server include a processor, hard disk, memory, system bus, etc. Servers are similar to general computer architectures, but because they need to provide highly reliable services, they have higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.
[0109] (5) Other electronic devices with data interaction functions.
[0110] This application also provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the foundation pit deformation prediction method described in any of the preceding embodiments.
[0111] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0112] The various embodiments in this specification are described in a related manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0113] For ease of description, if systems, servers, etc. are involved, they may be described separately as various units / modules based on their functions. Of course, in implementing this application, the functions of each unit / module can be implemented in one or more software and / or hardware.
[0114] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0115] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for predicting foundation pit deformation, characterized in that, The method includes: Acquire multi-source monitoring data of the foundation pit to be tested, and preprocess the data to construct a training dataset; A pre-built foundation pit deformation prediction model is trained based on the training dataset. The pre-built foundation pit deformation prediction model includes an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer. The input layer receives input features from the training dataset. The convolutional neural network module extracts local features from the input features and outputs a feature vector. The bidirectional long short-term memory network receives the feature vector, captures the long-term and short-term dependencies of foundation pit deformation over time, and outputs a hidden state sequence. The attention module performs adaptive weighted calculation on the hidden state sequence to generate a weighted feature vector. The output layer outputs predicted deformation data by flattening the weighted feature vector and then performing a non-linear transformation through a fully connected layer. The collected real-time monitoring data is acquired and input into the pre-built foundation pit deformation prediction model to obtain the predicted deformation data of the foundation pit to be tested.
2. The method for predicting foundation pit deformation according to claim 1, characterized in that, The process of acquiring multi-source monitoring data of the foundation pit to be tested and preprocessing it to construct a training dataset includes: Multi-source monitoring data of the foundation pit to be tested are acquired, and the multi-source monitoring data are analyzed based on the grey relational analysis method to obtain key feature data; the multi-source monitoring data includes historical deformation state variables, environmental factors, geological parameters, and construction factors; The key feature data is subjected to polynomial interpolation to obtain the processed key feature data; The processed key feature data is normalized to construct the original dataset; The original dataset is reconstructed using the sliding window method. Monitoring data from consecutive historical moments are used as input features, and deformation data from the next moment is used as prediction labels to form several training samples and construct a training dataset.
3. The method for predicting foundation pit deformation according to claim 2, characterized in that, The step of training a pre-built foundation pit deformation prediction model based on the training dataset includes: The training samples in the training dataset are input into the pre-built foundation pit deformation prediction model, and the input features are extracted locally based on the convolutional neural network module to obtain a feature vector; The feature vector is subjected to temporal evolution based on the bidirectional long short-term memory network module, and the hidden state sequence is output. Based on the attention module, an adaptive weighted calculation is performed on the hidden state sequence to generate a weighted vector; The weighted vector is forward propagated to calculate the error between the predicted and actual values of the foundation pit deformation. The model parameters are then iteratively updated using a backpropagation algorithm and an adaptive distance estimation optimization algorithm until the preset convergence condition is met, thus obtaining a pre-built foundation pit deformation prediction model that has been trained.
4. The method for predicting foundation pit deformation according to claim 2, characterized in that, The historical deformation state variables include at least the axial force of the supports and the horizontal displacement of the wall; the environmental factors include at least the groundwater volume, rainfall, and temperature; the geological parameters include at least the cohesion, internal friction angle, horizontal permeability coefficient, and average compression modulus of the soil layer; and the construction factors include at least the excavation depth of the foundation pit and the number of cross braces.
5. The method for predicting foundation pit deformation according to claim 1, characterized in that, The method further includes: The predicted deformation data is compared with a preset warning threshold. If the predicted deformation data exceeds the preset warning threshold, a corresponding warning signal is triggered.
6. A foundation pit deformation prediction system, characterized in that, The system includes: a dataset construction module, used to acquire multi-source monitoring data of the foundation pit to be tested, and preprocess it to construct a training dataset; The model training module is used to train a pre-constructed foundation pit deformation prediction model based on the training dataset. The pre-constructed foundation pit deformation prediction model includes an input layer, a convolutional neural network module, a bidirectional long short-term memory network module, an attention module, and an output layer. The input layer receives input features from the training dataset. The convolutional neural network module extracts local features from the input features and outputs a feature vector. The bidirectional long short-term memory network receives the feature vector, captures the long-term and short-term dependencies of foundation pit deformation over time, and outputs a hidden state sequence. The attention module performs adaptive weighted calculations on the hidden state sequence to generate a weighted feature vector. The output layer outputs predicted deformation data by flattening the weighted feature vector and then performing a non-linear transformation through a fully connected layer. The foundation pit deformation prediction module is used to acquire real-time monitoring data and input the real-time monitoring data into the pre-built foundation pit deformation prediction model to obtain the predicted deformation data of the foundation pit to be tested.
7. The foundation pit deformation prediction system according to claim 6, characterized in that, The dataset construction module includes: The data acquisition submodule is used to acquire multi-source monitoring data of the foundation pit to be tested, and analyze the multi-source monitoring data based on the grey relational analysis method to obtain key feature data; the multi-source monitoring data includes historical deformation state variables, environmental factors, geological parameters and construction factors; The first processing submodule is used to perform polynomial interpolation on the key feature data to obtain processed key feature data. The second processing submodule is used to normalize the processed key feature data and construct the original dataset. The dataset construction submodule is used to reconstruct samples from the original dataset using the sliding window method. It takes the monitoring data from consecutive historical moments as input features and the deformation data from the next moment as prediction labels to form several training samples and construct a training dataset.
8. The foundation pit deformation prediction system according to claim 6, characterized in that, The model training module is specifically used to input the training samples in the training dataset into the pre-built foundation pit deformation prediction model, and to extract local features from the input features based on the convolutional neural network module to obtain a feature vector; The feature vector is subjected to temporal evolution based on the bidirectional long short-term memory network module, and the hidden state sequence is output. Based on the attention module, an adaptive weighted calculation is performed on the hidden state sequence to generate a weighted vector; The weighted vector is forward propagated to calculate the error between the predicted and actual values of the foundation pit deformation. The model parameters are then iteratively updated using a backpropagation algorithm and an adaptive distance estimation optimization algorithm until the preset convergence condition is met, thus obtaining a pre-built foundation pit deformation prediction model that has been trained.
9. The foundation pit deformation prediction system according to claim 7, characterized in that, The historical deformation state variables include at least the axial force of the supports and the horizontal displacement of the wall; the environmental factors include at least the groundwater volume, rainfall, and temperature; the geological parameters include at least the cohesion, internal friction angle, horizontal permeability coefficient, and average compression modulus of the soil layer; and the construction factors include at least the excavation depth of the foundation pit and the number of cross braces.
10. The foundation pit deformation prediction system according to claim 6, characterized in that, The system also includes: The early warning module is used to compare the predicted deformation data with a preset early warning threshold. If the predicted deformation data exceeds the preset early warning threshold, a corresponding early warning signal is triggered.