PCA-PANN model-based slope stability prediction method and application
Through the PCA-PANN model, combined with principal component analysis and particle swarm optimization algorithm, a slope stability prediction model was constructed, which solved the efficiency and accuracy of slope stability prediction in the existing technology, and achieved a fast and accurate slope. Stability assessment provides efficient scientific decision-making support.
Patent Information
- Application Number
- CN202410754227.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-12
- Publication Date
- 2024-11-01
AI Technical Summary
The existing slope stability prediction methods have limitations, making it difficult to quickly and accurately analyze a large number of slopes in a limited time, especially in the rapid detection and early warning of landslide activities, and the numerical simulation method relies on complex engineering experience and long-term Time calculation makes it difficult to meet engineering needs.
Using the slope stability prediction method based on PCA-PANN model, the dimensionality reduction of principal component analysis, combined with particle swarm optimization algorithm and artificial neural network, a hybrid model is constructed to predict safety coefficients and identify the implicit relationship between input parameters and safety coefficients. Nonlinear relationships enable fast and accurate slope stability prediction.
This method significantly improves the computational efficiency, can process data without relying on measurement scales, deeply discover the implicit nonlinear relationship between variables, provides fast response and scientific decision support, and has a prediction accuracy of more than 0.97, which has significant advantages. Traditional method.
Smart Images

Figure CN118886293A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of slope stability prediction, and in particular to a slope stability prediction method based on a PCA-PANN model and an application thereof. Background Art
[0002] Slope is a common geological environment and load carrier in engineering construction. Figure 1 As shown in Figure 1, a large number of slopes will be formed during the construction of mines, roads, railways, water conservancy projects, and construction projects. Stability analysis is the most important part of any slope design study. Slope instability will cause huge losses to economic construction and people's lives. Therefore, building a reliable and effective slope stability prediction model has important theoretical and practical significance.
[0003] In order to evaluate the stability of the slope, a safety factor based on a suitable geotechnical model is required. The term safety factor (FoS) is an important indicator for evaluating the stability of the slope. Generally speaking, a slope with a safety factor (FoS) greater than 1.20 is considered safe. Since Terzaghi published his classic work "The Mechanism of Landslides", many researchers have contributed to a better understanding of slope stability and FoS. Traditional slope stability analysis methods mainly include limit equilibrium method and qualitative evaluation method. The limit equilibrium method was first proposed by Ferenius and is the earliest and most widely used method in slope stability research. The principle of the limit equilibrium method is to start from the force balance and calculate the FoS to evaluate the slope stability by calculating the relationship between the shear strength of the slope sliding body and the sliding force. Commonly used limit equilibrium methods include the Swedish arc method, the simplified Bishop method, the Spencer method, the Sarma method, etc. In addition, Zhou used the pseudo-dynamic method to establish a strict LEM considering the inter-band force, automatically search the sliding surface of the three-dimensional landslide and determine its safety factor. Recently, the safety factor of slopes in the Darjeeling-Kalimpong Himalaya region under dry and saturated conditions was studied using GeoStudio commercial software SLOPE / W based on the limit equilibrium method. In addition to LEM, a series of excellent numerical analysis methods such as finite element method (FEM), finite difference method (FDM), and FLAC have emerged to effectively solve complex slope stability problems. A large number of studies have combined numerical analysis methods to calculate the safety factor of critical sliding surfaces, quantitatively revealing the mechanism and process of gradual failure of rock and soil before and after instability.
[0004] The above studies have greatly deepened the understanding of slope instability. However, the slope is an open, uncertain, nonlinear and complex system. Affected by a variety of random and uncertain factors, the application of the above methods has great limitations and it is difficult to achieve ideal prediction results. For example, due to the existence of a large number of potential sliding surfaces, it is troublesome and difficult to find the critical sliding surface using the limit equilibrium method. Therefore, when a large number of slopes need to be analyzed in a limited time, it is not feasible to use LEM, especially for the rapid detection of landslide activities and the issuance of landslide warnings. The accuracy of the numerical simulation method depends on the setting of the constitutive model, boundary conditions and mechanical parameters, which often requires rich engineering experience and on-site back analysis to obtain reasonable results. Moreover, compared with the limit equilibrium method, the numerical simulation method takes longer. Therefore, a method that is easy to develop is needed to replace LEM and NA to achieve the purpose of rapid prediction of FS.
[0005] Thanks to the booming development of soft technology, the emergence of machine learning models has provided a bright future for addressing the challenges of Fos prediction. Based on the collected slope examples, applying machine learning methods to predict FoS has become an important solution. As a robust soft computing model, neural networks are widely used to evaluate slope stability. Neural networks include artificial neural networks, back propagation neural networks, differential evolution neural networks and other methods. Feng (2000) first proposed the concept of intelligent rock mechanics and proposed an intelligent slope stability prediction model based on neural networks. Bui et al. introduced deep learning neural networks in landslide assessment in Khun Tho Province, Vietnam, and compared its prediction performance with other classical machine learning algorithms. Foong and Moayedi used two new optimization techniques (i.e., vortex search algorithm and equilibrium optimization) to fine-tune the artificial neural network model for predicting FOS of single-layer slopes. Table A1 in the Appendix summarizes the typical neural network-based methods published in the past. So far, the vast majority of existing studies on slope stability prediction have adopted a single artificial intelligence technology, and more complex algorithms are still needed to improve slope stability prediction. Despite many efforts, there is still controversy about which model can achieve the most stable landslide displacement prediction. In addition, most machine learning models are validated using only a few landslide cases, a strategy that may lead to unreliable conclusions. As the requirements for FOS prediction accuracy in engineering projects become higher and higher, developing more potential ML models has become a trend.
[0006] The present invention maps high-dimensional raw data to low-dimensional data through principal component analysis and matrix transformation. Then, a hybrid model is developed to predict FOS by combining the advantages of particle swarm optimization and artificial neural network. Statistical indicators are introduced to evaluate the generalization performance of the model and the application of the established model is further discussed.
[0007] The occurrence of slope instability is a long-term, nonlinear and complex process. Due to the complexity of slope structure, the discontinuity of mechanical properties of geotechnical materials, and the variability of control factors acting on the slope, slope stability is affected by the combined influence of geological and engineering factors. As an indispensable key mechanical parameter in slope engineering design and stability analysis, FoS has been included in various standards and engineering specifications, and its research significance cannot be ignored. FoS evaluation involves many variables and requires the calculation of slope geometry data, geological material parameters, and pore water pressure. These uncertain factors have different weights on slope stability and have complex nonlinear relationships with each other.
[0008] The potential economic benefits of solving FoS through ML models are huge (e.g. allowing steeper slopes, better support system design, etc.). In practice, it is usually necessary to estimate FOS in the preliminary stages of engineering projects. When there are sufficient historical cases, data-driven techniques provide alternative models for slope stability assessment by establishing input-output relationships between safety factors and related parameters.
[0009] To this end, the present invention proposes a FoS intelligent prediction method in combination with the PCA method, and uses machine learning methods to train and develop a new prediction model for FoS. The input-output relationship between the safety factor and related parameters is identified, and the potential explicit or implicit relationship function is determined to predict the safety factor for a given set of input parameters. The new method can process data without relying on the measurement scale, and the arrangement of the data is not affected. In this model, the PSO algorithm is used to improve the performance of the ANN model. The present invention is a preliminary exploration of the PCA-PANN model in the field of slope stability research. Compared with traditional safety factors and numerical simulation methods, the model of the present invention can deeply discover and process implicit nonlinear relationships between variables. Therefore, it is a very promising slope stability prediction method. Summary of the invention
[0010] The purpose of the present invention is to provide a slope stability prediction method based on the PCA-PANN model, and introduce the PSO algorithm to optimize the ANN model. By inputting the information and mechanical parameters of the slope and combining the powerful ability of the machine learning model, an ideal and fast slope prediction result can be obtained.
[0011] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:
[0012] A slope stability prediction method based on PCA-PANN model includes the following steps:
[0013] S1: Collect factors related to slope stability as input variables and compile a database based on existing literature measurement data for training and developing ML models to predict FoS, including cohesion (c), internal friction angle Unit weight (γ), slope height (H), slope angle (Φ), pore pressure ratio (ru).
[0014] S2: Eliminate the correlation between the influencing factors in the database established in step S1 through principal component analysis (PCA) to establish new prediction sample data;
[0015] S3: Build and train the ANN model, which consists of interconnected neurons and builds a multi-layer neural network. Unit weight (γ), slope height (H), slope angle (Φ), and pore pressure ratio (ru) are transmitted and trained as input variables through neuron connections and activation functions. During the training process, the optimal number of neurons in the hidden layer is continuously adjusted to minimize the error between the predicted value and the actual value.
[0016] S4: Optimize ANN using particle swarm optimization (PSO) algorithm.
[0017] Furthermore, step S1 further includes: randomly dividing the entire data set into two independent subsets: a training set and a test set. The training set is used to build and train the model, and the test set is used to verify the prediction performance of the ML model. 80% of the entire data set is included in the training set, and the remaining 20% is included in the test set.
[0018] Furthermore, the step S2 specifically includes:
[0019] The original database X can be represented as a matrix with m rows and n columns, where m is the number of samples and n is the number of variables:
[0020]
[0021] The large difference in dimensions between the indicators will affect the prediction results. In order to better use the principal component analysis method, the original data is standardized and converted into a standardized matrix A according to formula 5.
[0022]
[0023] The correlation coefficient matrix R is calculated from the standardized matrix as shown in Formula 6
[0024] R m×n =X *T X * (6)
[0025] According to |λ-R|=0, the eigenvalue λ of the standardized data matrix is calculated, the eigenvalue is found, and finally the contribution rate of the principal component is obtained. The formula is as follows:
[0026]
[0027] The principal component analysis method is used to analyze the data in the database. The data standardized according to Formula 5 is analyzed, and finally the eigenvalues and cumulative contribution rates of the six principal components are obtained. According to the condition that the cumulative contribution rate of the principal component variance accounts for more than 80% of the total variance, the first four principal components are selected to replace the original variables for analysis.
[0028] Furthermore, the step S3 specifically includes: the ANN model learns weights associated with all neurons. As shown in Formula 8, the weighted sum of all inputs is transmitted through a nonlinear activation function.
[0029]
[0030] Among them, b and w i denotes the bias and respective weights and n denotes the number of inputs to the node.
[0031] Further, the step S4 specifically includes: assuming that there are N particles in the D-dimensional space, using X and V to represent the position and velocity of the particles, the position of the i-th particle is the velocity of the i-th particle as shown in formula 9, and the optimal position searched by the group, that is, the group optimal solution, is: Then the update formulas for the particle velocity and position are respectively:
[0032] X iD =(x i1 ,x i2 ,...,x iD ) (9)
[0033] The velocity of the i-th particle is shown in Equation 10:
[0034] V iD =(v i1 ,v i2 ,...,v iD ) (10)
[0035] In each iteration of the system, the particles move to find the best position (i.e., P best and G best ); the speed and position of all particles can be achieved:
[0036] V iD new =V iD +R1+C1R1(P best -X iD )+C2R2(G best-X iD ) (11)
[0037] X iD new =X iD +V iD new (12)
[0038] Among them, V iD new and X iD new Represents the updated speed and position of the particle; C1 and C2 are two custom position acceleration constants; R1 and R2 are two positive acceleration constants with a range of (0, 1).
[0039] On the other hand, the present invention proposes the application of the above method in the prediction of FoS of slope engineering.
[0040] Beneficial effects of the present invention:
[0041] Based on matrix transformation, the high-dimensional raw data can be mapped to low dimensions, and a hybrid model containing particle swarm algorithm and artificial neural network is constructed, which simplifies the original complex calculation model and greatly improves the calculation efficiency. This model algorithm collects and selects the main influencing variables of slope stability based on strength reduction, introduces evaluation indicators according to the optimization results, measures the effect of the prediction model, and finally derives sensitive factors according to characteristic parameters, and proposes slope protection measures for different sensitive factors.
[0042] 1. Data dimensionality reduction and information extraction:
[0043] Principal Component Analysis (PCA): PCA transforms the original data into a set of linearly unrelated variables (principal components) through linear transformation. These principal components are arranged in descending order of variance, so that the first few principal components can contain most of the information. This data dimensionality reduction technology not only reduces the input dimension of the model, but also effectively filters noise, improves the efficiency of data processing and the calculation speed of the model.
[0044] Information maximization: During the dimensionality reduction process, PCA preserves the main variability in the data as much as possible, which is very important for subsequent model training because these main variabilities often contain the information that has the greatest impact on the output variables (such as FoS).
[0045] 2. Optimize algorithms to improve model performance:
[0046] Particle Swarm Optimization (PSO): PSO is a swarm intelligence optimization algorithm that simulates the social behavior of bird flocks to find the optimal solution. During the ANN training process, PSO is used to optimize network weights and biases, effectively avoiding the local optimal problem that may be encountered in traditional training methods such as gradient descent, thereby improving the global search capability and convergence speed of the model.
[0047] Dynamic adjustment: The PSO algorithm continuously updates the speed and position of particles during the iteration process, and dynamically adjusts the search direction and step size according to the individual historical optimal position and the global optimal position. This enables the algorithm to find a good balance between exploration and utilization, speeding up convergence while maintaining good global search capabilities.
[0048] 3. Nonlinear modeling and complex relationship capture:
[0049] Artificial Neural Network (ANN): ANN can simulate the processing mechanism of the human brain and effectively capture and model the complex nonlinear relationship between input variables through multi-layer structure and nonlinear activation function. This is particularly important for problems such as slope stability, which are affected by multiple factors and have complex relationships, and can improve the accuracy of the model's prediction of the dynamic behavior of complex systems.
[0050] 4. Practical application and economic benefits:
[0051] Rapid response and decision support: By integrating PCA, PSO and ANN, the PCA-PANN model not only improves the accuracy of prediction, but also can quickly generate results, providing timely scientific basis for engineering decision-making, especially for rapid assessment of slope stability in emergency situations, which helps to take preventive measures in time and reduce possible economic losses and casualties.
[0052] The model of the present invention does not require complex calculations, and only needs to provide relevant characteristic parameters to quickly obtain the safety factor and stability state of the slope. In addition, the prediction accuracy of the PCA-PANN model for slope stability analysis reaches more than 0.97. Comparison with other neural network models shows that the model of the present invention has significant advantages in slope stability evaluation.
[0053] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0055] Figure 1 is a schematic diagram of the slope;
[0056] Figure 2 is the violin plot of the input variables;
[0057] Figure 3 This is a schematic diagram of the correlation matrix of the data set;
[0058] Figure 4 Schematic diagram of K-CV, (K=10);
[0059] Figure 5 Schematic diagram of the proposed ANN model structure;
[0060] Figure 6 Geometry diagram for particle velocity and position updates;
[0061] Figure 7 Schematic diagram of the PCA-PANN hybrid model framework;
[0062] Figure 8 Schematic diagram of the comparison between the measured and predicted values of the PANN model;
[0063] Fig. 9 Schematic diagram for comparing the measured and predicted values of the PCA-PANN hybrid model. DETAILED DESCRIPTION
[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0065] Example 1
[0066] The slope stability prediction method based on the PCA-PANN model described in this embodiment includes the following steps:
[0067] S1: Collect factors related to slope stability as input variables and compile a database based on existing literature measurement data for training and developing ML models to predict FoS, including cohesion (c), internal friction angle Unit weight (γ), slope height (H), slope angle (Φ), pore pressure ratio (ru). Among them, γ, c and It reflects the basic physical and mechanical indicators of rock and soil, ru reflects the size of pore water pressure, and H and Φ reflect the geometric characteristics of the slope.
[0068] S2: Eliminate the correlation between the influencing factors in the database established in step S1 through principal component analysis (PCA) to establish new prediction sample data;
[0069] S3: Build and train the ANN model, which consists of interconnected neurons and builds a multi-layer neural network. Unit weight (γ), slope height (H), slope angle (Φ), and pore pressure ratio (ru) are transmitted and trained as input variables through neuron connections and activation functions. During the training process, the optimal number of neurons in the hidden layer is continuously adjusted to minimize the error between the predicted value and the actual value.
[0070] S4: Optimize ANN using particle swarm optimization (PSO) algorithm.
[0071] In this embodiment, step S1 further includes: randomly dividing the entire data set into two independent subsets: a training set and a test set. The training set is used to build and train the model, and the test set is used to verify the predictive performance of the ML model. 80% of the entire data set is included in the training set, and the remaining 20% is included in the test set.
[0072] In this embodiment, step S2 specifically includes:
[0073] Slope instability presents typical volatility and multi-scale characteristics. Affected by random factors such as noise and interference, it is difficult to fully extract useful information from the data by directly modeling the original data, and the prediction accuracy is low. More importantly, there is a significant correlation between various types of data, which can easily cause repeated sample information input, increase the complexity of model training, and reduce generalization performance. PCA is a classic data dimensionality reduction and denoising method, which maps high-dimensional data into a few principal components in a low-dimensional space through matrix transformation. The original data is compressed for easy calculation, which improves the overfitting problem, and these principal components contain most of the useful information in the original data. The principal component analysis (PCA) is used to eliminate the correlation between the influencing factors in the database established in step S1, and establish new prediction sample data.
[0074] The original database X can be represented as a matrix with m rows and n columns, where m is the number of samples and n is the number of variables:
[0075]
[0076] The large difference in dimensions between the indicators will affect the prediction results. In order to better use the principal component analysis method, the original data is standardized and converted into a standardized matrix A according to formula 5.
[0077]
[0078] The correlation coefficient matrix R is calculated from the standardized matrix as shown in Formula 6
[0079] R m×n =X *T X * (6)
[0080] According to |λ-R|=0, the eigenvalue λ of the standardized data matrix is calculated, the eigenvalue is found, and finally the contribution rate of the principal component is obtained. The formula is as follows:
[0081]
[0082] The principal component analysis method is used to analyze the data in the database. The data standardized according to Formula 5 is analyzed, and finally the eigenvalues and cumulative contribution rates of the six principal components are obtained. According to the condition that the cumulative contribution rate of the principal component variance accounts for more than 80% of the total variance, the first four principal components are selected to replace the original variables for analysis.
[0083] In this embodiment, step S3 specifically includes:
[0084] Based on the function and structure of biological neurons, artificial neural network is a mathematical model that simulates the reasoning operation of the human brain and can solve complex mechanisms well. For multidimensional data with low correlation, ANN is the best alternative to traditional neural networks due to its excellent feature selection and function approximation capabilities. The basic unit of ANN is the neuron, which is responsible for receiving input features from the sensory organs and transmitting them to the brain. Neural networks contain three levels. The first layer is the input layer, through which all input data and parameters are fed into the neural network. The second layer is the hidden layer, which uses activation functions to process the input data. The processed data is finally output at the output layer.
[0085] The ANN model framework constructed by the present invention is as follows Figure 5 As shown in Figure 2, the model consists of interconnected neurons to build a multi-layer neural network. Input variables such as cohesion (c) and unit weight (γ) are transmitted and trained through neuron connections and activation functions. The optimal number of neurons in the hidden layer is continuously adjusted during the training process to minimize the error between the predicted value and the actual value. Training means that the ANN model needs to learn the weights associated with all neurons. The ANN model learns the weights associated with all neurons. As shown in Formula 8, the weighted sum of all inputs is passed through the nonlinear activation function.
[0086]
[0087] Among them, b and w i denotes the bias and respective weights and n denotes the number of inputs to the node.
[0088] In this embodiment, step S4 specifically includes:
[0089] The basic principle of the particle swarm algorithm is to use a group of particles with only speed and position attributes to simulate bird flight and predation. The location of the food represents the optimal solution to the problem, and the distance between the particle and the optimal solution represents the objective function value of the current particle. Each particle searches for the optimal position, and the optimal position it finds is the individual extreme value. Then each particle shares its individual extreme value with other particles, and finds the optimal individual extreme value through comparison. The optimal individual extreme value is the group optimal extreme value. Finally, all particles are updated according to the current individual optimal position and the group optimal position, and the above steps are repeated until the optimal solution is found.
[0090] Assume that there are N particles in D-dimensional space, and use X and V to represent the position and velocity of the particles. The position of the i-th particle is the velocity of the i-th particle as shown in formula 9; the optimal position searched by the group, that is, the group optimal solution, is: Then the update formulas for the particle velocity and position are:
[0091] X iD =(x i1 ,x i2 ,...,x iD ) (9)
[0092] The velocity of the i-th particle is shown in Equation 10:
[0093] V iD =(v i1 ,v i2 ,...,v iD ) (10)
[0094] In each iteration of the system, the particles move to find the best position (i.e., P best and G best ); the speed and position of all particles can be achieved:
[0095] V iD new=V iD +R1+C1R1(P best -X iD )+C2R2(G best -X iD ) (11)
[0096] X iD new =X iD +V iD new (12)
[0097] Among them, V iD new and XiD new Indicates the updated speed and position of the particle; C1 and C2 are two custom position acceleration constants; R1 and R2 are two positive acceleration constants with a range of (0, 1). In order to deepen the understanding of the particle swarm, the geometric diagram of the movement of particles in two-dimensional space is as follows Figure 6 shown.
[0098] The particle swarm optimization algorithm is used to search for the optimal number of neurons, learning rate, and number of iterations of the ANN. The weights and thresholds of the ANN network are mapped to a group of particles. The dimension of each particle represents the sum of the number of weights that play a connecting role in the network and the number of network thresholds. Using the PSO algorithm to train the neural network can give full play to its global optimization ability and local fast convergence advantages. The particles move and search in the weight and threshold space to minimize the error of the output layer of the neural network. Finally, a hybrid PANN model was developed by utilizing the advantages of the particle swarm algorithm and artificial neural network. The good global search ability of the particle swarm algorithm improves the generalization ability and learning performance of the neural network. The new model contains the advantages of PSO and ANN. Specifically, PSO searches for the global minimum in the search space, while ANN uses the determined global minimum to find the best result. In addition, the original data is subjected to principal component analysis in combination with PCA to extract new variables that meet the principal component requirements. The new variables are used as the input of the PANN model, and the final PCA-PANN flowchart is as follows. Figure 7 shown.
[0099] On the other hand, the present invention proposes the application of the above method in the prediction of FoS of slope engineering.
[0100] Example 2
[0101] Dataset preparation
[0102] The size of the dataset created from the experiments and field studies is limited. Due to the limitation of the publicly available geological data, it is not possible to comprehensively consider all potential factors. A database of 307 sets of data was compiled based on the available measurement data from the existing literature; it was used to train and develop the ML model to predict FoS.
[0103] The literature specifically includes:
[0104] 1. Wang J, Xu Y, Li J. Prediction of slope stability coefficient based on grid search supportvector machine. Railway Engineering. 2019; 59(05):94-7.
[0105] 2. Hong y, Shao z, Ma l.Application of a support vector machine for analysis and prediction of slope stability. Journal of Shenyang JianzhuUniversity.2017;33(06):1004-10.
[0106] 3. Su G. Fast estination of safety factor for circular failure rockslope using gaussian process model. Journal of Basic Science and Engineering. 2010; 18(6): 959-66.
[0107] 4. S, N, D.Anew approach to grid search method inslope stability analysis using Box–Behnken statistical design.Appl MathComput.2015;256:425-37.doi:https: / / doi.org / 10.1016 / j.amc.2015.01.022.
[0108] 5. XuX. Highway slope stability assessment based on the fuzzy-neural network [master] 2012.
[0109] 6. Wang C.Study on prediction methods for high engineering slope[master]2009.
[0110] Six most representative factors are selected to evaluate FoS. These six key factors include soil cohesion c, internal friction angle Soil weight γ, slope height H, slope angle Φ and pore pressure ratio ru. Among them, γ, c and It reflects the basic physical and mechanical indicators of rock and soil, ru reflects the size of pore water pressure, and H and Φ reflect the geometric characteristics of the slope.
[0111] Descriptive statistics are an important part of data analysis and provide a basis for subsequent ML modeling. Table 1 provides univariate descriptive statistics of the data, summarizing the data in an organized manner. The input parameters collected from different slopes have certain differences. Visual analysis can reflect the characteristic parameter information in the database and can perform qualitative analysis of the slope state to a certain extent. In order to evaluate whether there are outliers in the data, the present invention draws a violin plot of each characteristic parameter in the collected slope database. Figure 2 The violin plot combines the kernel density plot and the box plot to effectively depict the data distribution. The shape of the violin represents how the data is distributed along the vertical axis, and the interior of the violin contains a box plot. Figure 2 It can be seen that the cohesion is mostly distributed in the range of 10-36KPa; the slope angle is mainly 25°-45°. The slope height is mainly 40.3m. Figure 2 A rough inspection shows that the distribution of each input parameter in this database is relatively uniform. There are no outliers or few outliers in the data, indicating that the data set is reasonably constructed.
[0112] To study the correlation between characteristic parameters, Figure 3 Analyze the correlation between the factors affecting the data set and the output results. Figure 3 It can be found that two identical variables are represented by the symbol of the variable from the lower left to the upper right on the diagonal (representing a correlation coefficient of 1), and the correlation coefficients above the diagonal are symmetrical to those below the diagonal. The correlation coefficients between different variables are relatively low (most values are less than 0.4). In other words, it is not easy to find an obvious relationship between the six input variables and FoS, indicating that there is a strong nonlinear relationship between FoS and these six input variables.
[0113] Table 1 Statistical description of the data set used in the present invention
[0114]
[0115] In the subsequent machine learning modeling, the entire data set is randomly divided into two independent subsets: a training set and a test set. The training set is used to build and train the model, and the test set is used to verify the predictive performance of the ML model. The optimal percentage of the training set and the test set in the entire data set is determined by optimization analysis to obtain two subsets with sufficient representativeness. The present invention includes 80% of the entire data set in the training set and the remaining 20% in the test set.
[0116] Cross-validation and performance measurement
[0117] In order to minimize the deviation caused by the random division of training sets and test sets, the present invention adopts k-fold cross validation. K-fold cross validation is a method of dividing the database into smaller subsets in the statistical field to effectively eliminate the deviation caused by random sampling. The training set is divided into k groups of different subsets, each subset is used as a test set, and the remaining k-1 groups are used as new training sets. The accuracy of the ML model on each subset is evaluated to evaluate the effectiveness of the new method. Perform 10 or 5 cross validations. According to the size of the data set and the calculation time, the present invention adopts 10-fold cross validation. The process of 10-fold CV is as follows Figure 4 As shown in the figure, P1, P2, and P 10 They represent the prediction results of the corresponding folds respectively.
[0118] R 2 , mean absolute error (MAE) and root mean squared error (RMSE) have been widely used to evaluate the performance of ML models. The basic equations of each evaluation index are shown in equations (1)-(3)
[39] . Theoretically, when the performance result of the ML model is R 2 When =1, RMSE=0 and MAPE=0, the ML model performs best.
[0119]
[0120] Among them, FoS i pre and FoS i mea are the predicted results and the measured results respectively; E[FoS i mea ] indicates FoS i mea The average value of ; i = 1, 2…, N is the number of samples.
[0121] Example 3
[0122] The data in the database were analyzed according to the principal component analysis method. The data standardized according to equation 5 were analyzed, and the eigenvalues and cumulative contribution rates of the six principal components were finally obtained in Table 2. It can be seen that the cumulative contribution rate of the first four principal components is 85.036%, which meets the condition that the cumulative contribution rate of the principal component variance accounts for more than 80% of the total variance, and can fully reflect the main characteristics of the sample. Therefore, the first four principal components are selected to replace the original variables for analysis.
[0123] Table 2 Principal component eigenvalues and contribution rates
[0124]
[0125]
[0126] Table 3 Component matrix of normalized statistical parameters
[0127]
[0128] According to the factor score coefficient matrix (see Table 3), the expressions of each principal component can be obtained as follows:
[0129] As shown in formula 13:
[0130]
[0131] The superscript * represents the physical quantity after normalization in equation 5.
[0132] Model performance
[0133] For the sake of comparison, the effect of the PANN model without PCA processing is also shown. It is worth noting that before applying the PANN model, data preprocessing is also required to improve the prediction accuracy. The prediction effects of the two models are shown in Figure 2. Figure 8 As shown. Figure 8 As shown in Figure 2, the FoS value predicted by the PCA-PANN model is very close to the true value of FoS, whether in the training set or the test set. Figure 8 It can be seen that the FoS error generated by the PCA-PANN model is close to zero on most data sets, R 2 The prediction results of the PCA-PANN model are closest to the measured values, and the prediction error is the smallest (RMSE = 0.13, R 2 =0.971, MAE=0.125). The combination of machine learning and PCA enables the model to efficiently explore the most appropriate calculation parameters using principal component information, thereby improving the accuracy of FoS prediction.
[0134] Table 4 Comparison of prediction effects of various models
[0135]
[0136] In general, the new model has been shown to be effective in solving nonlinear FoS caused by multiple parameters, and the trained machine learning model can provide a reference for experienced team members. It is undeniable that, in terms of the current machine learning research results, the machine learning model cannot completely replace the traditional method to estimate FoS. However, if possible, based on the previous research data, using machine learning algorithms to accurately estimate the slope FoS with known design parameters will greatly reduce the test time and cost.
[0137] Example 4
[0138] In order to verify the prediction performance of the proposed method in actual engineering applications, the model of the present invention was applied to a slope project in Heihe City, Heilongjiang Province, China. The stability of this slope directly affects the stability of oil and gas transportation. The AA001-AA004 piles of the China-Russia East Line Natural Gas Pipeline are the control projects of the transit section. The slope angle is about 40°, the total length of the horizontal distance of the slope is 477m, and the height difference is about 120m. According to the on-site hydrogeological data (specific characteristic parameters are shown in Table 5), the literature (Bai G, Hou Y, Wan B, An N, Yan Y, Tang Z et al. Performance evaluation and engineering verification of machine learning based prediction models for slope stability. Appl Sci-Basel. 2022; 12 (15): 7890.) used a model established by numerical simulation method to evaluate the stability of the slope and calculated the FOS of the slope.
[0139] Table 5 Mechanical parameters of ramp examples
[0140]
[0141] Table 6 shows the FoS of the slope calculated by the ANN-method and the proposed PCA-PANN model based on the literature (Wang C. Study on prediction methods for high engineering slope [master] 2009). It can be seen that compared with the literature (Bai G, Hou Y, Wan B, An N, Yan Y, Tang Z et al. Performance evaluation and engineering verification of machine learning based prediction models for slope stability. Appl Sci-Basel. 2022; 12 (15): 7890.), the model of the present invention is closer to the numerical result. Interestingly, with the increase of water content, the FoS predicted by this model shows a trend of strengthening first and then weakening, which is consistent with the trend of slope stability obtained by numerical simulation method.
[0142] Table 6 Comparison of model prediction effects
[0143]
[0144] In summary, the present invention optimizes the ANN model by introducing the PSO algorithm. By inputting the slope information and mechanical parameters, combined with the powerful ability of the machine learning model, an ideal and fast slope prediction result can be obtained.
[0145] The present invention extracts four principal components from the slope influencing variables and eliminates the correlation between the influencing factors. The information volume carried by these principal components is 85.036%. The number of these principal components is less than that of the original variables. The PCA-PANN model is established with these principal components, and the model effect is significantly improved. The good statistical indicators also prove the effectiveness of the model in this paper.
[0146] The present invention compares the prediction results of artificial neural network with the FOS of a slope engineering example in Heihe City. The slope FoS prediction values obtained for each case are very close to the true value, proving the effectiveness and feasibility of the machine learning model.
[0147] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. A slope stability prediction method based on PCA-PANN model, characterized in that: The following steps are involved: S1: Collect factors related to slope stability as input variables and compile a database based on existing literature measurement data for training and developing ML models to predict FoS, including cohesion, internal friction angle, unit weight, slope height, slope angle, and pore pressure ratio; S2: Eliminate the correlation between the influencing factors in the database established in step S1 by principal component analysis, and establish new prediction sample data; S3: Build and train the ANN model, which consists of interconnected neurons and a multi-layer neural network. Cohesion, internal friction angle, unit weight, slope height, slope angle, and pore pressure ratio are used as input variables for transmission and training through neuron connections and activation functions. The optimal number of neurons in the hidden layer is continuously adjusted during the training process to minimize the error between the predicted value and the actual value. S4: Optimizing ANN using particle swarm optimization algorithm.
2. The slope stability prediction method based on the PCA-PANN model according to claim 1, characterized in that: The step S1 also includes: randomly dividing the entire data set into two independent subsets: a training set and a test set; the training set is used to build and train the model, and the test set is used to verify the predictive performance of the ML model; 80% of the entire data set is included in the training set, and the remaining 20% is included in the test set.
3. The slope stability prediction method based on the PCA-PANN model according to claim 1, characterized in that: The step S2 specifically includes: The original database X can be represented as a matrix with m rows and n columns, where m is the number of samples and n is the number of variables: The large difference in dimensions between the various indicators will affect the prediction results. In order to better use the principal component analysis method, the original data is standardized and converted into a standardized matrix A according to formula 5. The correlation coefficient matrix R is calculated from the standardized matrix as shown in Formula 6 R m×n =X *T X * (6) According to |λ-R|=0, the eigenvalue λ of the standardized data matrix is calculated, the eigenvalue is found, and finally the contribution rate of the principal component is obtained. The formula is as follows: The principal component analysis method is used to perform principal component analysis on the data in the database; the data standardized according to formula 5 is analyzed, and finally the eigenvalues and cumulative contribution rates of the six principal components are obtained. According to the condition that the cumulative contribution rate of the principal component variance accounts for more than 80% of the total variance, the first four principal components are selected to replace the original variables for analysis.
4. The slope stability prediction method based on the PCA-PANN model according to claim 1, characterized in that: The step S3 specifically includes: the ANN model learns weights associated with all neurons; as shown in Formula 8, the weighted sum of all inputs is passed through a nonlinear activation function; Among them, b and w i denotes the bias and respective weights and n denotes the number of inputs to the node.
5. The slope stability prediction method based on the PCA-PANN model according to claim 1, characterized in that: The step S4 specifically includes: assuming that there are N particles in a D-dimensional space, using X and V to represent the position and velocity of the particles, the position of the i-th particle is the velocity of the i-th particle as shown in formula 9, and the optimal position searched by the group, that is, the group optimal solution, is: Then the update formulas for the particle velocity and position are respectively: X iD =(x i1 ,x i2 ,...,x iD ) (9) The velocity of the i-th particle is shown in Equation 10: V iD =(v i1 ,v i2 ,...,v iD ) (10) In each iteration of the system, the particles move to find the best position, and the speed and position of all particles can reach: V iD new =V iD +R1+C1R1(P best -X iD )+C2R2(G best -X iD ) (11) X iD new =X iD +V iD new (12) Among them, V iD new and X iD new Represents the updated speed and position of the particle; C1 and C2 are two custom position acceleration constants; R1 and R2 are two positive acceleration constants with a range of (0, 1).
6. Application of the method according to any one of claims 1 to 5 in the prediction of FoS of slope engineering.
Citation Information
Cited By
Concrete interface shear strength prediction method, device and equipment and storage medium
CN119598578A
Slope stability probability prediction method and system and slope stability probability regression model construction system
CN120105931A
High slope chain generation risk assessment method based on numerical simulation
CN120805751A
Seabed slope failure probability evaluation and analysis method based on BP neural network algorithm
CN121351650A
