Slope support reliability assessment method and equipment

Abstract

The invention relates to the technical field of slope stability analysis, and particularly discloses a slope support reliability assessment method and equipment. According to the method, rock-soil body shear strength variability and supporting structure parameters are fully considered, an efficient and steady safety coefficient proxy model is trained and used for calculating slope safety coefficients under different supporting structures and rock-soil body parameters, and technical support is provided for rapid stability evaluation of slope engineering, slope supporting structure design and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of slope stability analysis, and in particular to a slope support reliability assessment method and equipment. Background Art

[0002] Slope disasters are one of the most common geological disasters, which seriously affect the safety of life and property and infrastructure construction. In order to evaluate the safety of slopes, the limit equilibrium method, finite element numerical simulation, strength reduction and other aspects are generally used to give the safety factor of the slope, but these deterministic analysis methods cannot consider the variability of rock and soil parameters. Due to the effects of load history conditions and physical and chemical processes, the shear strength parameters of the slope soil show spatial variability. In traditional slope engineering analysis, the shear strength parameters cannot be accurately determined through investigation, and the shear strength of the entire slope can only be represented by the local mean. The safety factor of the slope is given by deterministic analysis methods such as limit equilibrium method, finite element numerical simulation, and strength reduction. However, the existence of variability in rock and soil parameters will affect the credibility of deterministic analysis, and the existing analysis methods lack consideration of the uncertainty of rock and soil to evaluate whether the support structure is reliable. Summary of the invention

[0003] In order to overcome the impact of rock and soil parameter variability on the analysis results in the existing slope stability analysis, and the lack of consideration of the uncertainty of the rock and soil in the existing analysis method to evaluate whether the support structure is reliable, the present invention provides a slope support reliability assessment method and equipment.

[0004] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0005] In a first aspect, the present invention provides a slope support reliability assessment method, the method comprising:

[0006] A random field simulation is performed based on the geological body parameters of the target slope to generate a number of first shear strength distribution samples, and a random sampling is performed based on the range of the support parameters of the target slope to generate a number of support structure combination samples;

[0007] Randomly combining the plurality of first shear strength distribution samples and the plurality of support structure combination samples to generate a plurality of slope samples, and calculating a first safety factor of each slope sample through numerical simulation;

[0008] Extracting the geological body parameter distribution image, the geological body parameter average value and the support structure design parameters in the slope sample, and generating a data set according to the geological body parameter distribution image, the geological body parameter average value and the support structure design parameters corresponding to the first safety factor annotation;

[0009] Training a proxy model according to the data set to obtain a trained safety factor proxy model;

[0010] Based on the support parameter combination to be evaluated of the target slope, a random field simulation based on geological body parameters is performed on each support parameter combination to be evaluated to generate a number of second shear strength distribution samples;

[0011] Inputting each support parameter combination to be evaluated and the corresponding second shear strength distribution samples into the trained safety factor proxy model to obtain the second safety factors of each support parameter combination to be evaluated;

[0012] The reliability of each support parameter combination to be evaluated is calculated based on the plurality of second safety factors to obtain an evaluation result.

[0013] According to a specific implementation, in the above-mentioned evaluation method, the geological body parameters include the average cohesion value, the cohesion variation coefficient, the cohesion fluctuation distance, the friction angle average value, the friction angle variation coefficient and the friction angle fluctuation distance, the geological body parameter distribution image includes the cohesion distribution image and the friction angle distribution image, and the geological body parameter average value is the cohesion average value and the friction angle average value in the first shear strength distribution sample.

[0014] According to a specific implementation, in the above evaluation method, extracting the geological body parameter distribution image in the slope sample includes:

[0015] The cohesion value and friction angle value of each unit body in the first shear strength distribution sample are derived respectively to obtain the cohesion spatial distribution and the friction angle spatial distribution, and the value of each unit body in the cohesion spatial distribution and the friction angle spatial distribution is converted into grayscale intensity respectively to generate a cohesion distribution image and a friction angle distribution image.

[0016] According to a specific implementation, in the above-mentioned evaluation method, the safety factor proxy model includes a convolutional neural network and a fully connected neural network; wherein the convolutional neural network is used to extract features in the geological body parameter distribution image and output one-dimensional features; the one-dimensional features, the average values ​​of the geological body parameters and the support structure design parameters are spliced ​​to form a fusion feature, which is then input into the fully connected neural network to output the safety factor.

[0017] According to a specific implementation, in the above-mentioned evaluation method, the convolutional neural network includes a channel attention module, a spatial attention module, two standard convolutional layers, two maximum pooling layers and an expansion layer, the channel attention module includes a global pooling layer and a fully connected layer, and the spatial attention module includes a pooling layer and a one-dimensional convolutional layer.

[0018] According to a specific implementation, in the above evaluation method, the convolutional neural network is used to extract features in the geological body parameter distribution image and output two-dimensional features, specifically including:

[0019] After the geological body parameter distribution image is input into the first standard convolution layer, it is input into the channel attention module and the spatial attention module for feature selective processing, so as to improve the weights of the key channels and key areas of the image input features;

[0020] The processed image input features are input to the first maximum pooling layer, the second standard convolution layer, and the second maximum pooling layer, and then output one-dimensional features through the expansion layer.

[0021] According to a specific embodiment, in the above-mentioned evaluation method, the support parameters include the support parameters of the anchor rod or anchor cable and the anti-sliding pile parameters, and the support parameter combination to be evaluated includes the length of the anchor rod or anchor cable, the incident angle of the anchor rod or anchor cable, the prestress of the anchor rod or anchor cable, the spacing of the anchor rod or anchor cable and the length of the anti-sliding pile.

[0022] According to a specific implementation, in the above evaluation method, the reliability of each support parameter combination to be evaluated is calculated according to the plurality of second safety factors, specifically including:

[0023] The proportion of the second safety factors less than the preset threshold among the plurality of second safety factors is calculated to obtain the failure probability, and the formula is:

[0024] P f =P(F s <α)=m / M;

[0025] The reliability of each support parameter combination to be evaluated is calculated based on the failure probability, and the formula is:

[0026] P s =1-P f ;

[0027] Among them, P s is the reliability, P f For each support parameter combination to be evaluated, the failure probability, P(F s <α) is the second safety factor F s The probability of being less than the preset threshold α, m is the number of the second safety factor less than the preset threshold, and M is the number of the second safety factor.

[0028] In a second aspect, the present invention provides an electronic device, characterized in that it comprises at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a slope support reliability assessment method as described in any one of the above items.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] The present invention fully considers the variability of shear strength of rock and soil and the parameters of support structure, and trains an efficient and robust slope stability analysis agent model to calculate the slope safety factor under different support structures and rock and soil parameters, providing technical support for rapid stability assessment of slope engineering and design of slope support structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of a process flow of a slope support reliability assessment method provided by an embodiment of the present invention;

[0032] Figure 2 A schematic diagram of slope sample generation provided by an embodiment of the present invention;

[0033] Figure 3 A schematic diagram of a process flow of a slope support reliability assessment method provided by an embodiment of the present invention;

[0034] Figure 4 A schematic diagram of the structure of a safety factor proxy model provided by an embodiment of the present invention;

[0035] Figure 5 A schematic diagram of an image processing flow provided by an embodiment of the present invention;

[0036] Figure 6 A schematic diagram comparing the prediction effects of the safety factor proxy model provided by an embodiment of the present invention and the general model;

[0037] Figure 7 A schematic diagram of support parameters to be optimized provided in an embodiment of the present invention;

[0038] Figure 8 A schematic diagram of the calculation results of the safety factor of a sample provided by an embodiment of the present invention;

[0039] Fig. 9 A schematic diagram of performance evaluation of a safety factor proxy model provided by an embodiment of the present invention;

[0040] Fig.10 A schematic diagram of slope reliability assessment results for different support parameter combinations provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The present invention is further described in detail below in conjunction with test examples and specific implementation methods. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

[0042] The terms "first", "second", etc. in the embodiments of the specification, claims, and drawings of the present invention are used only for the purpose of distinguishing descriptions, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying an order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, including a series of steps or units. The method, system, product, or device is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0043] In addition, any embodiment or design described as "exemplary" or "for example" in the embodiments of the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way for easy understanding.

[0044] The present invention is mainly used to solve the problem of computational efficiency of slope reliability analysis considering the variability of shear strength of rock and soil, the problem that the purely data-driven machine learning agent model of the slope cannot consider the physical properties of the geological body, and the problem that the general attention mechanism cannot capture the complex structure and multi-level characteristics in the slope.

[0045] Based on the above problems, random reliability analysis can be used to estimate the reliability or failure probability of the slope from the uncertain strength parameters. The basic idea is to introduce the variability of the distribution of geotechnical parameters into the slope model through random field simulation and Monte Carlo sampling, and finally use numerical simulation methods such as finite element and finite difference to calculate the safety factor and reliability of the slope, so as to evaluate whether the support structure design meets the requirements under the uncertain slope geological body. The calculation accuracy of the reliability requires a large number of Monte Carlo sampling as support, so we are required to perform millions of numerical simulations, but these numerical simulations usually take a lot of time and computing power. In order to improve the computational efficiency, scholars use proxy models to predict the slope safety factor. For example, using models such as neural networks and support vector regression to construct the correlation between slope parameters and slope safety factors, the calculation amount of slope reliability analysis can be greatly reduced. Using methods such as neural networks and support vector regression, the correlation between simple parameters of the slope (such as slope height, slope angle, mean shear strength, etc.) and the safety factor can be constructed, that is, a purely data-driven proxy model. Purely data-driven machine learning methods cannot reflect the variability of shear strength parameters, and require a large number of training samples to improve model performance. Especially when there are complex support structures, pure data-driven models ignore the physical mechanism between the support structure and the slope rock and soil, resulting in poor generalization ability and poor interpretability. Therefore, it is necessary to develop a more robust slope stability analysis proxy model that can consider the spatial variability of parameters.

[0046] The slope support reliability assessment method provided by an embodiment of the present invention is introduced and illustrated in detail below.

[0047] Please refer to Figure 1 , which shows a schematic flow chart of a slope support reliability assessment method provided by an embodiment of the present invention, the method comprising:

[0048] Step 1: Perform random field simulation based on the geological parameters of the target slope to generate a number of first shear strength distribution samples, and perform random sampling based on the range of support parameters of the target slope to generate a number of support structure combination samples;

[0049] Step 2: Randomly combine the plurality of first shear strength distribution samples and the plurality of support structure combination samples to generate a plurality of slope samples, and calculate the first safety factor of each slope sample through numerical simulation;

[0050] Step 3: extracting the geological body parameter distribution image, the geological body parameter average value and the support structure design parameters in the slope sample, and generating a data set according to the geological body parameter distribution image, the geological body parameter average value and the support structure design parameters corresponding to the first safety factor annotation;

[0051] Step 4: training a proxy model according to the data set to obtain a trained safety factor proxy model;

[0052] Step 5: Based on the support parameter combination to be evaluated of the target slope, a random field simulation based on geological body parameters is performed on each support parameter combination to be evaluated to generate a number of second shear strength distribution samples;

[0053] Step 6: Input each support parameter combination to be evaluated and the corresponding second shear strength distribution samples into the trained safety factor proxy model to obtain a number of second safety factors for each support parameter combination to be evaluated;

[0054] Step 7: Calculate the reliability of each support parameter combination to be evaluated based on the plurality of second safety factors to obtain an evaluation result.

[0055] The embodiment of the present invention fully considers the variability of shear strength of rock and soil and the parameters of support structure, and trains an efficient and robust slope stability analysis agent model for calculating the slope safety factor under different support structures and rock and soil parameters, providing technical support for rapid safety assessment of slope engineering and design of slope support structures.

[0056] Specifically, in step 1, refer to Figure 2 , which shows a schematic diagram of sample generation provided by an embodiment of the present invention. Firstly, according to the refined finite difference numerical model of the typical cross section of the slope, according to the geological survey report and the support design parameters to be evaluated, the numerical simulation of the slope under different working conditions is carried out.

[0057] For geological body parameters, based on the average value, coefficient of variation, fluctuation distance and other parameters of cohesion and friction angle of different soil layers, the shear strength distribution of rock and soil body is simulated by random field simulation method to generate several first shear strength distribution samples.

[0058] For support parameters, the support parameters include support parameters of anchor rods or anchor cables and anti-sliding pile parameters, and the support parameter combination to be evaluated includes the length of the anchor rods or anchor cables, the incident angle of the anchor rods or anchor cables, the prestress of the anchor rods or anchor cables, the spacing of the anchor rods or anchor cables, and the length of the anti-sliding piles. After determining the range of the support parameters of the anchor rods or anchor cables and the anti-sliding pile parameters, the design parameters are randomly sampled by uniform sampling to generate several support structure combination samples.

[0059] According to the above steps, samples are generated and randomly combined to generate several slope samples. Each slope sample has a different shear strength distribution and support structure combination. The safety factor is calculated by the strength reduction method of numerical simulation to calculate the first safety factor of each slope sample.

[0060] Furthermore, the cohesion distribution image and the friction angle distribution image in each slope sample are extracted and exported and converted into grayscale images as image features. The average cohesion, average friction angle and support structure design parameters in each first shear strength distribution sample are taken as numerical features. A convolutional neural network with a multi-level attention mechanism is used to extract image features, and the processed image features are spliced ​​with numerical features such as rock and soil parameters and support structure parameters to form fusion features. The fusion features are processed by a fully connected neural network, and finally the safety factor corresponding to each slope sample is output. The safety factor label of the data set is the first safety factor calculated by numerical simulation in step 1. According to the geological body parameter distribution image, the average geological body parameter and the support structure design parameters corresponding to the first safety factor label, a data set is generated; the proxy model is trained according to the data set to establish a safety factor proxy model of physical information machine learning that can accurately predict the slope safety factor.

[0061] Furthermore, by inputting different support structure combination parameters and random field simulation results into the proxy model, the reliability of the slope under different parameter combinations can be quickly evaluated to assist in the design of slope support structures. Figure 3 , which shows a schematic flow chart of a slope support reliability assessment method provided by an embodiment of the present invention.

[0062] Specifically, the safety factor proxy model of physical information machine learning proposed in the embodiment of the present invention is mainly composed of a convolutional neural network including a multi-level attention mechanism and a fully connected neural network. In the safety factor proxy model, the distribution images of cohesion and friction angle are obtained from numerical simulation and belong to the category of physical information. At the same time, features such as the mean cohesion, the mean friction angle, and slope support structure parameters (such as pile length, anchor length, anchor incident angle, etc.) enhance the performance of the model from a data-driven perspective. Such a feature combination not only utilizes the physical information brought by numerical simulation, but also improves the prediction ability of the model by introducing key statistical parameters and design parameters.

[0063] Specifically, the cohesion distribution image and the friction angle distribution image are used as physical information and are preprocessed to form a dual-channel grayscale image. The dual-channel grayscale image is first passed through a standard convolution layer as an input feature, and then the feature is input into the multi-level attention module for feature selectivity processing to increase the weights of key channels and key areas of the image features. The image features are processed by the multi-level attention module and output to the first maximum pooling layer. This is followed by the second convolution layer and the second maximum pooling layer. The output of the second maximum pooling layer is expanded into one dimension by the expansion layer and connected with the numerical input features (shear strength parameters of the rock and soil mass, support structure design parameters, etc.) to form a complete fusion feature. The spliced ​​feature vector is output as a neural network composed of multiple layers of fully connected layers. The layers are connected by nonlinear activation functions, and the safety factor of the slope is finally output. Please refer to Figure 4 , which shows a schematic diagram of the structure of the safety factor proxy model provided by an embodiment of the present invention.

[0064] Among them, the color image is composed of three RGB channels, and the grayscale image is composed of one grayscale channel. The input image used by the proxy model proposed in the embodiment of the present invention is a grayscale image. The grayscale intensity value of each pixel in the grayscale image represents the color depth of the pixel at that point. According to this characteristic, the embodiment of the present invention obtains the spatial distribution of cohesion and the spatial distribution of friction angle by respectively deriving the cohesion value and the friction angle value of each unit body in the first shear strength distribution sample, and converts the value of each unit body in the cohesion spatial distribution and the friction angle spatial distribution into grayscale intensity, generates a cohesion distribution image and a friction angle distribution image, and then inputs them into the convolutional neural network as features. Specifically, the slope attribute view is opened in the numerical analysis software, and the color of each area in the numerical model displayed in the window represents the attribute value of the position. For example, when the cohesion attribute view is opened, different grid points have different colors, and from blue to red, the cohesion is from small to large, which is very similar to the grayscale intensity value of the pixel point of the grayscale image.

[0065] Therefore, the embodiment of the present invention processes the cohesion distribution image and the friction angle distribution image according to the following method:

[0066] (1) Random field discretization: First, random field simulation is performed based on the geological parameters of the target slope to obtain its shear strength distribution, and the cohesion and friction angle values ​​of each unit body are derived respectively, that is, the spatial distribution of cohesion and friction angle of the sample is obtained;

[0067] (2) Mapping to digital image: The spatial distribution of cohesion and friction angle is converted into a grayscale digital image. Specifically, the coordinates of the numerical model unit correspond to the image pixel coordinates, and the shear strength value (cohesion or friction angle) of the unit is converted into a grayscale intensity (0-255).

[0068] (3) Image generation: Create two grayscale images, representing the distribution of cohesion or friction angle, respectively;

[0069] (4) Additional numerical features: Calculate the average value of the cohesion or friction angle achieved by each random field and use it as an additional numerical input feature.

[0070] This method enables the present invention to utilize the powerful capabilities of convolutional neural networks in processing images, thereby effectively capturing the spatial variability of slope soil.

[0071] Furthermore, the input includes two grayscale digital images (representing the distribution of cohesion or friction angle, respectively) and two numerical features (mean values ​​of cohesion and friction angle) as well as the design parameters of the support structure. The processing flow is as follows: Figure 5 shown.

[0072] It is understandable that the attention module is an important part of the convolutional neural network architecture design. Similar to the selective processing of complex scenes by the human visual system, the convolutional neural network adopts the attention mechanism to focus on specific parts of the input features. It enables the convolutional neural network to focus on key areas or channels of key images, reduce unnecessary calculations and improve efficiency.

[0073] In order to enhance the performance of the proxy model in the prediction of slope safety factor, the present invention uses a multi-level attention mechanism to process the cohesion distribution image and the friction angle distribution image to enhance the model's ability to identify key areas such as potential sliding surfaces.

[0074] Specifically, in a possible implementation, the input feature map is assumed to be Where H, W and C represent the height, width and number of channels of the feature map respectively. The input feature map consists of two channels, which are grayscale images representing the cohesion distribution and friction angle distribution of the slope physical model.

[0075] (1) Channel Attention Module

[0076] Step 1: Global Pooling

[0077] The features are first passed through a global pooling layer to identify which shear strength parameter (cohesion or friction angle) has a greater impact on the slope stability. First, global average pooling and global maximum pooling are performed on the input feature map F:

[0078]

[0079]

[0080] Among them, the global average pooling and global maximum pooling of the kth channel can be expressed as:

[0081]

[0082]

[0083] After global average pooling, the results of all channels are summarized into a tensor F with a dimension of 1×1×C avg , after global maximum pooling, the results of all channels are summarized into a tensor F with a dimension of 1×1×C max .

[0084] Step 2: Weight Generation

[0085] The feature vector after global maximum pooling and average pooling is input into a shared fully connected layer. This fully connected layer is used to learn the attention weights of each channel. Through learning, the network can adaptively decide which channels are more important for the current task.

[0086] For global average pooling:

[0087]

[0088] For global max pooling:

[0089]

[0090] in and is the weight matrix of the fully connected layer, r is the compression ratio, σ is the activation function, and the ReLU activation function is used for nonlinear transformation. To ensure that the attention weights are between 0 and 1, the Sigmoid activation function is applied to generate the channel attention weights. The weight matrices W1 and W2 are initialized when defining the fully connected layer. Throughout the training process, as each batch of data is input and back-propagated, the weight matrices W1 and W2 are gradually adjusted to minimize the loss function. After multiple training rounds (epochs), these weight matrices will gradually converge to values ​​that best capture the important channels in the input features.

[0091] Step 3: Feature Weighting

[0092] Multiply the channel attention weights by the original feature map element by element to enhance the feature expression of the key channels:

[0093]

[0094] Where ⊙ represents element-wise multiplication. Multiply the attention weights by each channel of the original feature map to obtain the attention-weighted channel feature map. This will emphasize channels that are helpful for the current task and suppress irrelevant channels.

[0095] (2) Spatial Attention Module

[0096] Step 1: Pooling

[0097] The feature map after channel weighting is in F channel Average pooling and maximum pooling are performed on the channel dimension. This operation averages the values ​​of all channels at each pixel position to generate an output feature map containing a single channel.

[0098]

[0099]

[0100] Average all channel values ​​at each spatial position (i, j)(i, j)(i, j):

[0101]

[0102]

[0103] Step 2: Weight Generation

[0104] The pooled feature maps are concatenated in the channel dimension, and a two-dimensional convolutional layer (Conv2D) is used to process the concatenated feature maps, and a two-dimensional convolutional layer with 1 filter is applied. This operation integrates the information from the concatenated channels and outputs a single-channel attention weight map to generate a spatial attention weight Wspatial, which is used to adjust the weights of each spatial position of the input feature map.

[0105]

[0106]

[0107] Step 3: Feature Weighting

[0108] Multiply the original feature map by the spatial attention weights element-wise:

[0109]

[0110] The final output is the feature map F after channel and spatial attention weighting spatial , which is used for further slope safety factor prediction analysis.

[0111] The multi-attention mechanism combines the channel and spatial attention mechanisms to extract and amplify the key information of the feature map in the channel and spatial dimensions, thereby improving the ability to identify important shear strength parameters and potential sliding areas of the slope.

[0112] Specifically, the slope support reliability assessment method proposed in the embodiment of the present invention uses a machine learning neural network based on physical information as a proxy model for slope safety factor calculation, reducing the amount of calculation for numerical simulation; the shear strength distribution image of the rock and soil body is used as one of the input features of the model, which can characterize the physical properties of the rock and soil body, and incorporates the constraints of physical characteristics on the basis of pure data drive. The multi-level attention mechanism is adopted, and the model has a significant improvement in identifying key slope geological body areas and key shear strength parameters, providing higher accuracy and stronger generalization ability.

[0113] The safety factor proxy model based on physical information machine learning proposed in the embodiment of the present invention shows excellent performance in slope safety factor prediction and reliability assessment, and has obvious advantages over traditional methods:

[0114] The computational efficiency is greatly improved: By introducing a proxy model based on physical information, the present invention greatly reduces the amount of numerical calculations required in the traditional Monte Carlo simulation method. In scenarios where millions of numerical simulations are usually required, the proxy model of the present invention can achieve the same computational accuracy with only a small amount of numerical simulation calculations through an efficient machine learning algorithm. This will greatly shorten the evaluation time in slope engineering design and provide strong support for the rapid optimization of complex slope design.

[0115] Fully consider the spatial variability of rock and soil: This paper inputs the spatial variability of the shear strength of rock and soil into the model through physical information images, overcoming the limitation that the traditional pure data-driven model cannot reflect the physical characteristics of the geological body. By inputting the distribution of shear strength into the model in the form of grayscale images and combining the multi-level attention mechanism for feature extraction, the model's ability to recognize complex slope structures and potential sliding surfaces is significantly enhanced, making the safety factor prediction more accurate.

[0116] Improve the generalization ability and accuracy of the model: Compared with traditional proxy models such as neural networks or support vector machines, the present invention combines physical information and geological design parameters and uses a multi-level attention mechanism to effectively capture the key areas and parameters in complex slopes.

[0117] Better interpretability: By introducing the physical information of shear strength, the present invention improves the interpretability of the model. The model output can not only directly predict the safety factor of the slope, but also be used to analyze the impact of support design parameters on the slope reliability. This combination of physics and data greatly enhances the reliability and acceptability of the model in actual engineering applications.

[0118] It should be noted that the interpretability of the surrogate model is crucial in engineering applications because it helps engineers understand the model's prediction mechanism and make necessary adjustments. Grad-CAM images can be used to qualitatively evaluate the interpretability of the surrogate model in predicting the safety factor. Grad-CAM is a visualization method that clearly shows the model's focus area by using the importance of spatial positions in the gradient calculation convolution layer.

[0119] In order to verify the improvement of the safety factor proxy model provided by the embodiment of the present invention in terms of interpretability and accuracy, the Grad-CAM image of the safety factor predicted by the safety factor proxy model is compared with the general model (LeNet5 convolutional neural network model without using the multiple attention mechanism), as shown in FIG. Figure 6 As shown. The two models were trained and tested on the same dataset. Six samples (including three failed samples) were randomly selected from the validation dataset, and Grad-CAM was used to generate a heat map of the feature weights of the last convolutional layer output to show the feature areas that the model focuses on when predicting the safety factor of the slope. At the same time, we also derived the maximum shear strain increment map when calculating the safety factor using numerical simulation. The shear strain increment represents the degree of deformation in each area. Areas with large shear strain increments may indicate that significant shear deformation has occurred in the area, so the local area of ​​the shear strain increment can characterize the potential sliding surface. The heat map generated by Grad-CAM was qualitatively compared with the maximum shear strain increment map. The safety factor proxy model provided by the embodiment of the present invention always focuses on areas with large shear strain increments, which indicate potential slope instability, indicating that the safety factor proxy model provided by the embodiment of the present invention is better than the general convolutional neural network model in identifying key areas. And the safety factor values ​​predicted by this model are more aligned with the values ​​calculated by numerical simulation, especially in samples with a safety factor less than 1. Therefore, the safety factor proxy model provided by the embodiment of the present invention significantly enhances the model's ability to identify key features of the slope and improves the performance and interpretability of the proxy model.

[0120] In summary, the present invention not only improves the efficiency and accuracy of slope stability analysis, but also provides a more intelligent decision support tool for slope engineering design, which has engineering application value and market potential.

[0121] The technical solution provided by the present invention is further described below in conjunction with specific implementation cases.

[0122] Specifically, the present invention intends to verify the slope safety factor under different design parameters while considering the parameter variability, so as to provide a reference for the slope support design. Therefore, in this embodiment, a refined finite difference numerical model of a typical cross section of the slope is first established, and numerical simulation of the slope under different working conditions is carried out according to the geological survey report and the support design parameters to be evaluated. For the slope geological body parameters, based on the average values ​​of cohesion and friction angle, coefficient of variation, fluctuation distance and other parameters of different soil layers, the shear strength distribution of the rock and soil body is simulated by the random field simulation method.

[0123] The design parameters to be optimized include anchor length, anchor angle, anchor prestress, anchor spacing, anti-slide pile length, etc. Figure 7 Specifically, in a possible implementation, the support parameters are set to include: anchor length L = 15-60m, anchor incident angle α = 20-40°, anchor spacing (out-of-plane) s = 2.5-6m, anchor prestress F = 600-800kN, and anti-sliding pile length D = 15-30m.

[0124] After determining the range of design parameters, the design parameters are randomly selected by uniform sampling. In summary, the python-flac3d interface is used to realize the automation process, randomly generate 500 groups of slope samples with random working conditions, and calculate the safety factor by the strength reduction method of numerical simulation. The safety factor calculation results of 500 groups of slope samples obey the normal distribution as follows Figure 8 The rock and soil parameters and support structure parameters are shown in the following table.

[0125] Table 1 is an example table of rock and soil parameters

[0126]

[0127] Table 2 is an example table of support parameters

[0128] Support parameters Anchor length Anchor incident angle Anchor prestressing Anchor spacing Anti-slide pile length Value range 15-60m 20-40° 600-800kN 2.5～6m 15～30m

[0129] The 500 groups of slope samples obtained according to the above steps are used as training samples for the proxy model, and the safety factor proxy model is obtained after training. The 10,000 groups of labeled samples that did not participate in the training are input into the proxy model, and the error between the safety factor output by the proxy model and the safety factor calculated by numerical simulation (as the true value) is compared to evaluate the performance of the model. Fig. 9As shown), the model performs well on the validation set, with a variance of 0.00961 and a mean absolute deviation of 0.07739 for the safety factor prediction. The above-mentioned support parameters to be evaluated are used to generate 100 different combinations of support parameters to be evaluated, and each combination of support parameters to be evaluated is subjected to a random field simulation based on geological body parameters to obtain 500 groups of second shear strength distribution samples. Based on the trained safety factor proxy model, each combination of support parameters to be evaluated and the corresponding second shear strength distribution samples are simultaneously input to obtain 500 second safety factors corresponding to 500 groups of second shear strength distribution samples for each combination of support parameters to be evaluated, and then the reliability of each combination of support parameters to be evaluated is calculated based on the several second safety factors. Specifically, if m of the M second safety factors are less than a preset threshold, then when M is large enough, the law of large numbers shows that the failure probability P and reliability index of the slope are:

[0130] P f =P(F s <α)=m / M,P s =1-P f ;

[0131] Among them, P s is the reliability, P f For each support parameter combination to be evaluated, the failure probability, P(F s <α) is the second safety factor F s is less than the preset threshold α, m is the number of second safety factors less than the preset threshold, and M is the number of second safety factors. In this embodiment, for better explanation, the preset threshold is set to 1, and the slope reliability corresponding to the above 100 combinations of support parameters to be evaluated is plotted as follows: Fig.10 As shown, it can be seen that the slope reliability under parameter combination 53 is the highest.

[0132] It is understandable that the number of each sample or the number of combinations mentioned above should not be interpreted as being more preferred or more advantageous than other embodiments or design solutions. Specifically, the numbers mentioned above are intended to present the relevant concepts in a specific way for easy understanding.

[0133] On the other hand, an embodiment of the present invention further provides an electronic device, which includes a processor, a network interface and a memory, wherein the processor, the network interface and the memory are interconnected, wherein the memory is used to store a computer program, and the computer program includes program instructions, and the processor is configured to call the program instructions to execute the above-mentioned slope support reliability assessment method.

[0134] In the embodiments of the present invention, the processor may be an integrated circuit chip having the ability to process signals. The processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0135] The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The processor reads the information in the storage medium and completes the steps of the above method in combination with its hardware.

[0136] The storage medium may be a memory, which may be, for example, a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memory.

[0137] Among them, the non-volatile memory can be a read-only memory (ROM), a programmable ROM (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory.

[0138] The volatile memory may be a random access memory (RAM) which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), sync link DRAM (SLDRAM), and direct rambus RAM (DRRAM).

[0139] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0140] It should be understood that the system disclosed in the present invention can be implemented in other ways. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the communication connection between modules can be through some interfaces, indirect coupling or communication connection of servers or units, which can be electrical or other forms.

[0141] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each module may exist physically separately, or two or more modules may be integrated into one processing unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional unit.

[0142] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A slope support reliability assessment method, characterized in that: The method comprises: A random field simulation is performed based on the geological body parameters of the target slope to generate a number of first shear strength distribution samples, and a random sampling is performed based on the range of the support parameters of the target slope to generate a number of support structure combination samples; Randomly combining the plurality of first shear strength distribution samples and the plurality of support structure combination samples to generate a plurality of slope samples, and calculating a first safety factor of each slope sample through numerical simulation; Extracting the geological body parameter distribution image, the geological body parameter average value and the support structure design parameters in the slope sample, and generating a data set according to the geological body parameter distribution image, the geological body parameter average value and the support structure design parameters corresponding to the first safety factor annotation; Training a proxy model according to the data set to obtain a trained safety factor proxy model; Based on the support parameter combination to be evaluated of the target slope, a random field simulation based on geological body parameters is performed on each support parameter combination to be evaluated to generate a number of second shear strength distribution samples; Inputting each support parameter combination to be evaluated and the corresponding second shear strength distribution samples into the trained safety factor proxy model to obtain the second safety factors of each support parameter combination to be evaluated; The reliability of each support parameter combination to be evaluated is calculated based on the plurality of second safety factors to obtain an evaluation result.

2. A slope support reliability assessment method according to claim 1, characterized in that: The geological body parameters include an average cohesion value, a coefficient of cohesion variation, a cohesion fluctuation distance, an average friction angle value, a coefficient of friction angle variation and a friction angle fluctuation distance; the geological body parameter distribution image includes a cohesion distribution image and a friction angle distribution image; and the geological body parameter average values ​​are the average cohesion values ​​and the average friction angle values ​​in the first shear strength distribution sample.

3. A slope support reliability assessment method according to claim 2, characterized in that: Extracting a geological body parameter distribution image in the slope sample includes: The cohesion value and friction angle value of each unit body in the first shear strength distribution sample are derived respectively to obtain the cohesion spatial distribution and the friction angle spatial distribution, and the value of each unit body in the cohesion spatial distribution and the friction angle spatial distribution is converted into grayscale intensity respectively to generate a cohesion distribution image and a friction angle distribution image.

4. A slope support reliability assessment method according to claim 1, characterized in that: The safety factor proxy model includes a convolutional neural network and a fully connected neural network; wherein the convolutional neural network is used to extract features in the geological body parameter distribution image and output one-dimensional features; the one-dimensional features, the average values ​​of the geological body parameters and the support structure design parameters are spliced ​​to form a fusion feature, which is then input into the fully connected neural network to output the safety factor.

5. A slope support reliability assessment method according to claim 4, characterized in that: The convolutional neural network includes a channel attention module, a spatial attention module, two standard convolutional layers, two maximum pooling layers and an expansion layer. The channel attention module includes a global pooling layer and a fully connected layer. The spatial attention module includes a pooling layer and a two-dimensional convolutional layer.

6. A slope support reliability assessment method according to claim 5, characterized in that: The convolutional neural network is used to extract features from the geological body parameter distribution image and output one-dimensional features, specifically including: After the geological body parameter distribution image is input into the first standard convolution layer, it is input into the channel attention module and the spatial attention module for feature selective processing, so as to improve the weights of the key channels and key areas of the image input features; after the processed image input features are input into the first maximum pooling layer, the second standard convolution layer and the second maximum pooling layer, the one-dimensional features are output through the expansion layer.

7. A slope support reliability assessment method according to claim 1, characterized in that: The support parameters include support parameters of anchor rods or anchor cables and anti-sliding pile parameters, and the support parameter combination to be evaluated includes the length of the anchor rods or anchor cables, the incident angle of the anchor rods or anchor cables, the prestress of the anchor rods or anchor cables, the spacing of the anchor rods or anchor cables and the length of the anti-sliding piles.

8. A slope support reliability assessment method according to claim 1, characterized in that: The reliability of each support parameter combination to be evaluated is calculated according to the plurality of second safety factors, specifically including: The proportion of the second safety factors less than the preset threshold among the plurality of second safety factors is calculated to obtain the failure probability, and the formula is: P f =P(F s <α)=m / M; The reliability of each support parameter combination to be evaluated is calculated based on the failure probability, and the formula is: P s =1-P f ; Among them, P s is the reliability, P f For each support parameter combination to be evaluated, the failure probability, P(F s <α) is the second safety factor F s The probability of being less than the preset threshold α, m is the number of the second safety factor less than the preset threshold, and M is the number of the second safety factor.

9. An electronic device, characterized in that: It includes at least one processor and a memory that is communicatively connected to the at least one processor; the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a slope support reliability assessment method as described in any one of claims 1 to 8.

Images