RNN-based expansive soil slope freeze-thaw crack depth calculation method
Through the RNN-based deep learning method and combined with CNN for feature extraction, the problems of large manpower investment and poor accuracy of traditional detection methods are solved, and efficient and accurate detection of freeze-thaw crack depth on the slope of expansive soil is achieved.
Patent Information
- Application Number
- CN202510207598.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-25
- Publication Date
- 2025-06-13
AI Technical Summary
When detecting the depth of freeze-thaw cracks on the expansive soil slope, the prior art investment is large, the cost is high, the time is consumed, and the accuracy is poor, making it difficult to effectively evaluate the stability of the slope.
Using RNN-based calculation method, a freeze-thaw cycle experiment was performed on the expanded soil sample, combined with CNN for feature extraction, a deep learning model was established, and the weight parameters and bias parameters in the model were optimized by Adam optimizer to achieve accurate calculation of crack depth.
This method can effectively overcome the shortcomings of traditional detection methods, improve the accuracy and efficiency of crack depth detection on the slope of expansive soil, reduce labor investment and cost, and significantly improve the speed and accuracy of detection.
Smart Images
Figure CN120145002A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of crack identification and optimization, and in particular to a calculation method for the freeze-thaw crack depth of expansive soil slopes based on RNN. Background Art
[0002] Expansive soil is widely distributed in China. Generally, expansive soil has high bearing capacity, but it is highly sensitive to water and has characteristics such as water absorption and swelling, water loss and shrinkage, attenuation of bearing capacity when soaked in water, and development of dry shrinkage cracks. Its properties are extremely unstable. When the seasonal freezing depth exceeds 0.5 m, obvious freeze-thaw disasters will occur to engineering buildings. In particular, when expansive soil is under the alternating action of freeze-thaw cycles, it will cause the cracks in the soil mass to develop rapidly, resulting in a reduction in strength. A large number of engineering practices have shown that the freeze-thaw cycle is one of the important factors leading to the deformation and landslide failure of expansive soil slopes. As the distribution range of the seasonal frozen area gradually increases, the expansive soil mass within its range is repeatedly frozen and melted, affecting the soil structure and mechanical indexes. In the seasonal frozen soil area, due to the large temperature difference between winter and summer, the expansive soil slope project experiences repeated freeze-thaw cycles, which has a non-negligible impact on the slope stability.
[0003] The freeze-thaw cycle causes changes in the matrix suction of the soil mass in the frost heave area of the slope. When the tensile stress exceeds its own tensile capacity, the expansive soil mass undergoes uneven shrinkage and the number of cracks increases. The increase in cracks not only reduces the integrity of the soil mass, but also increases the permeability of the expansive soil, increases the degree of water infiltration, and significantly reduces the shear strength and safety factor of the expansive soil slope, resulting in a significant deterioration of the engineering properties and making it more likely to cause landslide accidents in the soil mass of the expansive soil slope. A large number of studies have shown that cracks play a controlling role in the strength of expansive soil. The final cracking depth of the cracks is closely related to the instability sliding depth of the expansive soil slope and is the key to the prediction and prevention of expansive soil engineering disasters. At the same time, the timely discovery and treatment of the problems of expansive soil slopes will greatly reduce the losses caused by their damage.
[0004] At present, most methods for detecting the crack depth of soil masses adopt manual detection means. The detection process usually involves on-site detection personnel going to the vicinity of the structure to be detected, using relevant detection equipment to measure the crack depth, and then manually recording relevant information such as the crack position and corresponding parameters. This type of method requires a large amount of manpower, high cost, long time consumption, and poor accuracy. Summary of the Invention
[0005] The present invention provides a calculation method for the depth of freeze-thaw cracks in expansive soil slopes based on RNN, which can effectively optimize the weight parameters and bias parameters in the model by applying the Adam optimizer, thereby achieving the purpose of minimizing the loss. At the same time, by training the model, an effective model can be obtained, and these models are used to evaluate their capabilities for comparison with other models. This method can overcome the defects of theoretical models that adopt multiple assumptions, have poor universality, and are difficult to calibrate advanced model parameters, and solves the problems raised in the above background technology.
[0006] The present invention provides the following technical solutions: A calculation method for the depth of freeze-thaw cracks in expansive soil slopes based on RNN, comprising the following steps:
[0007] Step S1: Conduct a freeze-thaw cycle experiment on the expansive soil sample and record the corresponding depth of freeze-thaw cracks;
[0008] Step S2: Extract features through CNN;
[0009] Step S3: Establish a deep learning model through RNN;
[0010] Step S4: Model training.
[0011] Preferably, step S1 specifically includes:
[0012] By burying graphite electrodes at both ends of the expansive soil sample and connecting wires, and connecting the wires to an electrochemical workstation, an electrical signal dataset of expansive soil samples with different crack depths is obtained.
[0013] Preferably, step S2 specifically includes:
[0014] One-dimensional convolutional neural network (1DCNN) is specifically used to process sequence data, so that 1DCNN extracts high-dimensional features in time or space from one-dimensional sequence data for effective data recognition and classification. Each convolutional layer applies multiple convolutional kernels to filter the input data, and the convolutional kernels capture local features in the data; the subsequent pooling layer is responsible for reducing the feature dimension and subsampling the features to reduce the computational amount and prevent overfitting.
[0015] Preferably, step S3 specifically includes:
[0016] Process variable-length sequences through RNN, and capture the temporal information in the sequences through RNN to achieve the capture of the relevance of context information when processing natural language, and capture the dependency relationships and semantic information between words.
[0017] Preferably, step S4 specifically includes:
[0018] a. Initialization of parameters: Set parameters such as weights and learning rates;
[0019] b. Establish the loss function and set the stopping condition: Use the mean squared error loss function to calculate the mean squared error between them. The calculation formula is shown below:
[0020]
[0021] where N is the length of the input sequence, is the predicted value of the model, and y i is the true value;
[0022] c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized;
[0023] d. According to the preset number of training times, continuously adjust the parameters of the model until a certain specific threshold is reached. When the expected number of training times is not reached, the model stops running when it meets the threshold. After 3 epochs, if the loss function has not improved significantly, the deep learning model will be terminated.
[0024] The present invention has the following beneficial effects:
[0025] 1. The calculation method for the freeze-thaw crack depth of expansive soil slopes based on RNN, the one-dimensional convolutional neural network (1DCNN), is specifically used to process sequence data, enabling the 1DCNN to extract high-dimensional features in terms of time or space from one-dimensional sequence data for effective data recognition and classification. Each convolutional layer applies multiple convolutional kernels to filter the input data, and the convolutional kernels are used to capture local features in the data; the subsequent pooling layer is responsible for reducing the feature dimension and subsampling the features to reduce the computational amount and prevent overfitting.
[0026] 2. The calculation method for the freeze-thaw crack depth of expansive soil slopes based on RNN processes sequence data through RNN. Different from traditional feedforward neural networks, RNN has recurrent connections, enabling it to maintain a memory state when processing sequences. In RNN, there is a hidden state at each time step, which can receive the input of the current time step and the hidden state of the previous time step as inputs. The output of the hidden state depends not only on the input of the current time step but also on the inputs of all previous time steps. By processing variable-length sequences through RNN and capturing the temporal information in the sequences, the relevant information of the context is captured when processing natural language, and the dependency relationships and semantic information between words are captured. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 is a schematic diagram of the process of the present invention;
[0028] Figure 2Schematic diagram of the 1DCNN unit structure of the present invention;
[0029] Figure 3 Schematic diagram of the RNN module structure of the present invention. Specific embodiments
[0030] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
[0031] Please refer to Figure 1 - Figure 3 , a calculation method for the depth of freeze-thaw cracks in expansive soil slopes based on RNN, comprising the following steps:
[0032] Step S1: Conduct a freeze-thaw cycle experiment on the expansive soil specimen and record the corresponding depth of freeze-thaw cracks;
[0033] Step S2: Extract features through CNN;
[0034] Step S3: Establish a deep learning model through RNN;
[0035] Step S4: Model training.
[0036] In a preferred embodiment, step S1 specifically includes:
[0037] By burying graphite electrodes at both ends of the expansive soil specimen and connecting wires, and connecting the wires to an electrochemical workstation, an electrical signal dataset of expansive soil specimens with different crack depths is obtained.
[0038] In a preferred embodiment, step S2 specifically includes:
[0039] One-dimensional convolutional neural network (1DCNN) is a variant of CNN, specifically designed for processing sequence data. It usually consists of an input layer, several convolutional layers, pooling layers, fully connected layers, and an output layer. The alternating use of convolutional layers and pooling layers not only increases the depth of the model but also reduces the data dimension while maintaining key information. This structural design makes 1DCNN particularly good at extracting high-dimensional features in time or space from one-dimensional sequence data for effective data recognition and classification. Each convolutional layer applies multiple convolutional kernels to filter the input data, and these convolutional kernels can capture local features in the data; the subsequent pooling layer is responsible for reducing the feature dimension and subsampling the features to reduce the computational amount and prevent overfitting.
[0040] In a preferred embodiment, step S3 specifically includes:
[0041] RNN (Recurrent Neural Network) is a type of recurrent neural network used to process sequential data. Different from traditional feedforward neural networks, RNNs have recurrent connections, enabling them to maintain a memory state when processing sequences. In an RNN, there is a hidden state at each time step, which can receive the input of the current time step and the hidden state of the previous time step as inputs. The output of the hidden state depends not only on the input of the current time step but also on the inputs of all previous time steps. By processing variable-length sequences with an RNN and capturing the temporal information in the sequences, the relevant information of the context can be captured when processing natural language, and the dependency relationships and semantic information between words can be captured.
[0042] In a preferred embodiment, step S4 specifically includes:
[0043] a. Initialization of parameters: Parameters such as weights and learning rates need to be set in order to be correctly applied to the model;
[0044] b. Establishing the loss function and setting the stopping condition: The loss function is an important learning criterion that can be used to measure the performance of a network and is often related to the optimization quantity. The mean squared error loss function is used to calculate the mean squared error between them, and its calculation formula is shown below:
[0045]
[0046] where N is the length of the input sequence, is the predicted value of the model, and y i is the true value;
[0047] c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized;
[0048] d. Through careful adjustment of the parameters and functions, an effective network training process can be achieved, that is, according to the preset number of training times, the parameters of the model are continuously adjusted until a certain specific threshold is reached. Even if the expected number of training times is not reached, the model can stop running when the threshold is met. After 3 epochs, if the loss function has not improved significantly, the deep learning model will be terminated.
[0049] Working principle: By embedding graphite electrodes at both ends of the expansive soil specimen and connecting wires, and connecting the wires to an electrochemical workstation, an electrical signal dataset of expansive soil specimens with different crack depths is obtained;
[0050] Then, one-dimensional convolutional neural network is used to process sequence data, which usually consists of an input layer, several convolutional layers, pooling layers, fully connected layers and an output layer. The alternating use of convolutional layers and pooling layers not only increases the depth of the model, but also reduces the data dimension while retaining key information. This structural design makes 1DCNN particularly good at extracting high-dimensional features in time or space from one-dimensional sequence data, so as to perform effective data recognition and classification. Each convolutional layer applies multiple convolutional kernels to filter the input data, and these convolutional kernels can capture local features in the data; the subsequent pooling layer is responsible for reducing the feature dimension and subsampling the features to reduce the computational amount and prevent overfitting;
[0051] Then, RNN is used to process sequence data. Different from traditional feedforward neural networks, RNN has recurrent connections, which enables it to maintain a memory state when processing sequences. In RNN, there is a hidden state at each time step, which can receive the input of the current time step and the hidden state of the previous time step as inputs. The output of the hidden state depends not only on the input of the current time step, but also on the inputs of all previous time steps. By using RNN to process variable-length sequences and capturing the temporal information in the sequences, the relevance of context information when processing natural language can be captured, and the dependency relationships and semantic information between words can be captured;
[0052] Finally, through the simulation training steps:
[0053] a. Initialization of parameters: These parameters such as weights and learning rate need to be set in order to be correctly applied to the model;
[0054] b. Establishment of loss function and setting of stopping conditions: The loss function is an important learning criterion, which can be used to measure the performance of a network and is often related to the optimization quantity. The mean square error loss function is used to calculate the mean square error between them, and its calculation formula is shown below:
[0055]
[0056] where N is the length of the input sequence, is the predicted value of the model, and y i is the true value;
[0057] c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized;
[0058] d. Through careful adjustment of parameters and functions, an effective network training process can be achieved, that is, according to the preset number of training times, continuously adjust the parameters of the model until a certain specific threshold is reached. Even if the expected number of training times is not reached, the model can stop running when the threshold is met. After 3 epochs, if the loss function has not improved significantly, the deep learning model will be terminated, thus overcoming the defects of the theoretical model, such as adopting multiple assumptions, poor universality, and difficulty in calibrating the parameters of the advanced model.
[0059] It should be noted that in this article, relational terms such as first and second are only used 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 term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.
[0060] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for calculating the depth of freeze-thaw cracks in expansive soil slopes based on RNN, characterized in that: The following steps are involved: Step S1: performing a freeze-thaw cycle experiment on an expansive soil sample and recording the corresponding freeze-thaw crack depth; Step S2: feature extraction through CNN; Step S3: Establish a deep learning model through RNN; Step S4: Model training.
2. The method for calculating the depth of freeze-thaw cracks of expansive soil slopes based on RNN according to claim 1, characterized in that: Step S1 specifically includes: By burying graphite electrodes at both ends of the expansive soil sample and connecting wires, and connecting the wires to the electrochemical workstation, the electrical signal data sets of the expansive soil samples with different crack depths were obtained.
3. The method for calculating the depth of freeze-thaw cracks of expansive soil slope based on RNN according to claim 1, characterized in that: Step S2 specifically includes: One-dimensional convolutional neural network (1DCNN) is specifically used to process sequence data, so that 1DCNN can extract high-dimensional features in time or space from one-dimensional sequence data for effective data recognition and classification. Each convolution layer applies multiple convolution kernels to filter the input data and uses the convolution kernels to capture local features in the data; the subsequent pooling layer is responsible for reducing the feature dimension and sub-sampling the features to reduce the amount of calculation and prevent overfitting.
4. The method for calculating the depth of freeze-thaw cracks of expansive soil slope based on RNN according to claim 1, characterized in that: Step S3 specifically includes: By processing variable-length sequences through RNN and capturing the timing information in the sequence, it is possible to capture the relevance of contextual information when processing natural language, and capture the dependency and semantic information between words.
5. The method for calculating the depth of freeze-thaw cracks of expansive soil slope based on RNN according to claim 1, characterized in that: Step S4 specifically includes: a. Parameter initialization: setting the parameters of weight and learning rate; b. Establish the loss function and set the stopping condition: Use the mean square error loss function to calculate the mean square error between them. The calculation formula is as follows: Where N is the length of the input sequence, is the predicted value of the model, y i is the true value; c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized; d. According to the preset number of training times, the model parameters are continuously adjusted until a certain threshold is reached. If the expected number of training times is not reached, the model stops running when the threshold is met. After 3 epochs, if the loss function has not been significantly improved, the deep learning model will be terminated.