Railway dangerous rock stability intelligent prediction method based on RF-GA-SVM model

Abstract

The invention discloses a railway dangerous rock stability intelligent prediction method based on an RF-GA-SVM model, and the method comprises the following steps: determining railway slope dangerous rock stability evaluation characteristics, and building a railway slope dangerous rock stability evaluation system; relevant data of railway slope dangerous rock, geology, weather and the like are collected, an initial data set is established, and data preprocessing is carried out; eliminating redundant evaluation features by using variable importance scores of a random forest algorithm; carrying out global optimization on the SVM model hyper-parameters by adopting a genetic algorithm; based on the SVM model after hyper-parameter optimization, a railway slope dangerous rock stability prediction model RF-GA-SVM is constructed; and training, testing and evaluating the RF-GA-SVM model to obtain a final railway slope dangerous rock stability prediction model. According to the method, the three algorithms of RF, GA and SVM are fused, the accuracy and efficiency of railway slope dangerous rock stability prediction are improved, the robustness and applicability of the prediction model are remarkably improved, and powerful technical support is provided for railway safety management and geological disaster prevention and control.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring and prediction of railway geological disasters, and in particular to an intelligent prediction method for the stability of dangerous rocks on railway slopes that integrates a random forest algorithm (RF), a genetic algorithm (GA) and a support vector machine (SVM). Background Art

[0002] At present, the intelligent prediction methods for the stability of dangerous rocks on railway slopes mainly include mechanical analysis, empirical judgment and mathematical model methods. Among them, the mechanical analysis method emphasizes the mechanical modeling of slope structures, but it has high requirements for parameter accuracy and calculation conditions; the empirical rule focuses on the subjective judgment of engineers and lacks universality; the mathematical model rule attempts to establish an evaluation function from the perspective of statistics or intelligent calculation, but it still has certain limitations in the face of the high-dimensional complexity of multi-source data. In recent years, machine learning methods have become a research hotspot, such as support vector machine (SVM), fuzzy C-means clustering (FCM), artificial neural network (ANN) and other models, which have been widely used in slope hazard classification. However, there are two key problems in existing research: (1) the selection of input features is highly subjective and the importance of each indicator cannot be scientifically evaluated; (2) the adjustment method of model parameters is rough and the optimization is insufficient, which affects the final prediction accuracy and model generalization ability. In particular, SVM is highly sensitive to parameters C and γ, and traditional grid search or manual experience is difficult to cope with complex engineering scenarios. Therefore, there is an urgent need for an intelligent model that integrates feature selection and parameter optimization mechanism, which can not only effectively extract key indicators but also improve the model learning ability and stability. Summary of the Invention

[0003] The present invention provides an intelligent prediction method for railway dangerous rock stability based on the RF-GA-SVM model, aiming to solve the problems and defects mentioned in the above background technology.

[0004] In order to achieve the above objectives, the following technical solutions are provided: An intelligent prediction method for railway dangerous rock stability based on RF-GA-SVM model includes the following steps: S1. Determine the characteristics of railway slope dangerous rock stability evaluation and establish a railway slope dangerous rock stability evaluation system; S2. Based on the characteristics of railway slope dangerous rock stability evaluation, collect railway slope dangerous rock, geology, and weather related data, establish an initial data set, and preprocess the initial data set; S3. Use the random forest algorithm to score the importance of each railway slope rock stability evaluation feature and remove redundant features based on the score; S4, using genetic algorithm to globally optimize the hyperparameters of the SVM model; S5. Based on the SVM model after hyperparameter optimization, the railway slope dangerous rock stability prediction model RF-GA-SVM was constructed; S6. Divide the preprocessed initial data set into a training set and a test set, and use the training set to train the RF-GA-SVM model; S7. The RF-GA-SVM model was tested using the test set and its performance was comprehensively evaluated to obtain the final railway slope dangerous rock stability prediction model.

[0005] Furthermore, in step S3, the importance scoring of each railway slope dangerous rock stability evaluation feature using the random forest algorithm includes the following steps: S31: Use the Bootstrap method to randomly select sample sets and construct a decision tree; S32: Randomly select a feature subset for the decision tree node and perform optimal splitting; S33: Calculate the importance of the stability evaluation features of each railway slope dangerous rock using the OOB error. The calculation formula is: ; Where, For the The importance of features, For the Out-of-bag errors, To disrupt the Out-of-bag error after features, is the number of trees in the random forest.

[0006] Furthermore, in step S4, the global optimization of the hyperparameters of the SVM model using a genetic algorithm includes the following steps: S41: parameter initialization, including population size, crossover probability, mutation probability and number of iterations; S42: Use the fitness function Fitness to evaluate the quality of individuals. The fitness calculation formula is: Where, For individuals The fitness of For the current individual The mean square error of the corresponding SVM model, It is a small positive number to avoid division by zero errors; S43: Perform a selection operation in a roulette-wheel manner to determine the probability of inheriting the next generation of individuals. The calculation formula is: , where For individuals The probability of being selected, pop is the population size; S44: Perform crossover and mutation operations, where: , where is the original value of a gene in the current chromosome before mutation, is the new value of a gene in the current chromosome after mutation, is a random number uniformly distributed in the interval [0,1] to ensure the randomness of the mutation operation. 、 are the upper and lower limits of the parameter values respectively; S45: When the convergence condition is met or the maximum number of iterations is reached, the iteration is stopped and the optimal parameters are output, including the penalty factor C and the kernel function width γ.

[0007] Furthermore, in step S5, the railway slope dangerous rock stability prediction model RF-GA-SVM is constructed based on the SVM model after hyperparameter optimization, including the following steps: The SVM model is defined using the RBF kernel function as: ; Where, is the input sample, For the input sample The predicted output value of is the number of training samples, is the RBF kernel function, is the sample feature vector, 、 are all Lagrange multipliers, express The weight in the decision function, In the standard classification SVM, it is always zero. is the bias term; The calculation formula of RBF kernel function is: , where is the kernel function parameter; The SVM model determines parameters by optimizing the following objective function: , where is the hyperplane normal vector, is the bias term, is the penalty factor, is the slack variable, is the number of training samples, and Constraints must be met: ,in, For the The label of each sample is +1 or -1, representing the positive and negative classes respectively.

[0008] Furthermore, the determination of the railway slope dangerous rock stability evaluation characteristics at least takes into account the slope dangerous rock conditions, rock mass structure characteristics, and hydrogeological conditions.

[0009] Furthermore, the preprocessing of the initial data in step S2 at least includes: filling in missing values, removing outliers and data standardization.

[0010] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention introduces the random forest (RF) algorithm to perform feature selection, automatically evaluates the importance of each railway slope dangerous rock stability evaluation feature, and screens out key features that have a significant impact on the prediction of railway slope dangerous rock stability. Compared with traditional empirical methods and subjective screening, RF feature selection can scientifically and objectively identify redundant features, reduce data dimensions, avoid model overfitting caused by redundant information, and greatly improve the prediction accuracy of the model; (2) The present invention uses a genetic algorithm (GA) to globally optimize the hyperparameters of the support vector machine (SVM). The genetic algorithm automatically finds the optimal parameters of the SVM (penalty factor C and kernel function width γ) by simulating the process of natural selection, effectively avoiding the subjectivity of manual parameter adjustment and the problem of local optimal solution in traditional methods. Through GA optimization, the parameter setting of the SVM model is more precise, thereby improving the classification accuracy and generalization ability of the model. (3) The present invention significantly improves the quality of input data and ensures the accuracy and completeness of the data set by systematically preprocessing the original data, including missing value filling, outlier removal and data standardization. The Z-score standardization method eliminates the dimensional differences between features, ensuring that the model can better adapt to the training of different features and avoiding the degradation of model performance due to data scale differences. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a flow chart of a railway dangerous rock stability prediction algorithm according to an embodiment of the present invention; Figure 2 This is a diagram of a railway dangerous rock stability evaluation system according to an embodiment of the present invention; Figure 3 This is the confusion matrix diagram of the RF-GA-SVM model classification in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0013] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0014] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0015] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0016] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0017] Reference Figure 1 , an intelligent prediction method for railway dangerous rock stability based on RF-GA-SVM model, the specific implementation method is as follows: Step 1: Based on railway engineering geological standards and relevant literature, determine the characteristics of railway slope dangerous rock stability evaluation and establish a railway slope dangerous rock stability evaluation system. The specific operations are as follows: like Figure 2 As shown in the figure, based on railway engineering geological standards and relevant literature, and taking into account factors such as the dangerous rock conditions on the slopes, rock mass structural characteristics and hydrogeological conditions, a railway slope dangerous rock stability evaluation system with 14 secondary evaluation features was established to ensure the scientific nature and wide adaptability of the evaluation results. By quantifying and standardizing different influencing factors, the shortcomings of subjective judgment and non-standardized evaluation in traditional methods can be effectively avoided, and a comprehensive and adaptable railway slope dangerous rock stability evaluation index system can be constructed.

[0018] Step 2: Based on the determined railway slope dangerous rock stability evaluation characteristics, collect relevant data such as railway slope dangerous rock, geology, and weather, establish an initial data set, and preprocess the initial data set. The specific operations are as follows: On-site data collection of dangerous rock formations on railway slopes was conducted, along with raw data including geological parameters, climatic conditions, and rock mechanical properties. This data was used to construct an initial dataset. Missing values, outliers, and redundant features in the initial dataset were first addressed using mean interpolation, and outliers were removed using the interquartile range method. Data standardization was then used to convert each indicator to zero mean and unit variance, enabling equal treatment of features of different dimensions within the same model, improving both accuracy and robustness.

[0019] Step 3: A random forest algorithm is used to evaluate the importance of each evaluation feature in the railway slope rock stability assessment system and to remove redundant features based on the scores. This step automatically identifies key features that significantly impact rock stability prediction. This process effectively reduces redundant variables, lowers the model's dimensionality, and improves computational efficiency. Furthermore, the RF model exhibits excellent noise immunity, enabling it to cope with noise and uncertainty in slope data. The main steps are as follows: S31: Use the Bootstrap method to randomly select sample sets and construct a decision tree; S32: Randomly select a feature subset for the decision tree node and perform optimal splitting; S33: Calculate the importance of the stability evaluation features of each railway slope dangerous rock using the OOB error. The calculation formula is: ; Where, For the The importance of features, For the Out-of-bag errors, To disrupt the Out-of-bag error after features, is the number of trees in the random forest.

[0020] Step 4: Use genetic algorithm to globally optimize the hyperparameters of the SVM model; Traditional support vector machines (SVMs) rely on kernel function parameters (such as the penalty factor C and kernel width γ) for classification when processing high-dimensional, nonlinear data. However, the selection of these parameters often relies on manual experience and can easily lead to local optimal solutions. Therefore, this paper uses a genetic algorithm (GA) to globally optimize SVM hyperparameters. By simulating the process of natural selection, the GA continuously selects, crosses over, and mutates parameters to ensure the global optimal solution is found, thereby improving the model's generalization ability and prediction accuracy. The main steps are as follows: S41: parameter initialization, including population size, crossover probability, mutation probability and number of iterations; S42: Use the fitness function Fitness to evaluate the quality of individuals. The fitness calculation formula is: Where, For individuals The fitness of For the current individual The mean square error of the corresponding SVM model, It is a small positive number to avoid division by zero errors; S43: Perform a selection operation in a roulette-wheel manner to determine the probability of inheriting the next generation of individuals. The calculation formula is: , where For individuals The probability of being selected, pop is the population size; S44: Perform crossover and mutation operations, where: , where is the original value of a gene in the current chromosome before mutation, is the new value of a gene in the current chromosome after mutation, is a random number uniformly distributed in the interval [0,1] to ensure the randomness of the mutation operation. 、 are the upper and lower limits of the parameter values respectively; S45: When the convergence condition is met or the maximum number of iterations is reached, the iteration is stopped and the optimal parameters are output, including the penalty factor C and the kernel function width γ.

[0021] Step 5: Based on the hyperparameter-optimized support vector machine (SVM) model, a railway slope rock stability prediction model (RF-GA-SVM) is constructed. By constructing a hyperplane in a high-dimensional space, the support vector machine can effectively classify "stable" and "unstable" rock formations on the slope. By using the RBF kernel function, the SVM can handle complex nonlinear relationships and adapt to the classification of multidimensional features in the geological environment. The SVM optimized by the genetic algorithm can provide higher classification accuracy and stronger adaptability when processing complex data. The following steps are included: The SVM model is defined using the RBF kernel function as: ; Where, is the input sample, For the input sample The predicted output value of is the number of training samples, is the RBF kernel function, used to calculate the input sample With support vector The inner product after nonlinear mapping; is the sample feature vector, 、 are all Lagrange multipliers, express The weight in the decision function, In the standard classification SVM, it is always zero. is the bias term; The calculation formula of RBF kernel function is: , where is the kernel function parameter; The SVM model determines parameters by optimizing the following objective function: , where is the hyperplane normal vector, is the bias term, is the penalty factor, is the slack variable, is the number of training samples, and Constraints must be met: ,in, For the The label of each sample is +1 or -1, representing the positive and negative classes respectively; if =0, indicating sample is completely correctly classified and outside the boundary; if 0< <1, description sample is correctly classified, but is between the boundary and the interval; if >1, description sample Misclassified.

[0022] Step 6: Divide the preprocessed dataset into training set and test set according to the proportion, and use the training set to train the RF-GA-SVM model; The present invention divides the preprocessed data into a training set and a test set in a ratio of 8:2. By training the training set with SVM model, combined with RF feature selection and GA hyperparameter optimization, it ensures that the model can show good prediction effect in actual data. like Figure 3 As shown in the figure, the confusion matrix output after the model training shows that the RF-GA-SVM model has good classification ability in the four types of dangerous rock stability levels. All prediction results are concentrated on the diagonal line, and there is no category confusion, which reflects the high efficiency of the model in sample learning, feature recognition and boundary division; especially in the multi-classification scenario, it maintains a high recognition rate, reflecting the superiority of the combination of RF feature extraction + GA optimized SVM kernel parameters.

[0023] Step 7: The RF-GA-SVM model is tested using the test set and its performance is comprehensively evaluated to obtain the final railway slope dangerous rock stability prediction model. In this embodiment, when verified on the test set, the RF-GA-SVM model is able to predict the stability level of railway slope dangerous rock, and compared with the actual survey results, the accuracy rate reaches over 96%; At the same time, to comprehensively evaluate the performance of the proposed RF-GA-SVM model, the present invention conducted comparative experiments with a traditional support vector machine (SVM) model and a GA-SVM model optimized solely using a genetic algorithm. The experimental results are shown in Table 1. The RF-GA-SVM model outperformed the control model in terms of classification accuracy, error control, and training convergence speed. For example, the RF-GA-SVM model achieved a classification accuracy of 96.67%, while the traditional SVM model achieved an accuracy of 88.34%. Furthermore, RF-GA-SVM demonstrated significant advantages in terms of training convergence speed and model error.

[0024] Table 1: Performance comparison of the RF-GA-SVM model according to the present invention, the traditional SVM model, and the GA-SVM model optimized by the genetic algorithm in the task of predicting the stability of dangerous railway rocks

[0025] 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 scope of protection of the present invention.

Claims

1. An intelligent prediction method for railway dangerous rock stability based on RF-GA-SVM model, characterized by: The following steps are involved: S1. Determine the characteristics of railway slope dangerous rock stability evaluation and establish a railway slope dangerous rock stability evaluation system; S2. Based on the railway slope dangerous rock stability evaluation characteristics, collect railway slope dangerous rock, geology, and weather related data to establish an initial data set, and preprocess the initial data set; S3. Use the random forest algorithm to score the importance of each railway slope rock stability evaluation feature and remove redundant features based on the score; S4, using genetic algorithm to globally optimize the hyperparameters of the SVM model; S5. Based on the SVM model after hyperparameter optimization, the railway slope dangerous rock stability prediction model RF-GA-SVM was constructed; S6. Divide the preprocessed initial data set into a training set and a test set, and use the training set to train the RF-GA-SVM model; S7. The RF-GA-SVM model was tested using the test set and its performance was comprehensively evaluated to obtain the final railway slope dangerous rock stability prediction model.

2. The intelligent prediction method for railway dangerous rock stability according to claim 1 is characterized in that: In step S3, the importance of evaluating the stability of dangerous rock masses on each railway slope is scored using the random forest algorithm, including the following steps: S31: Use the Bootstrap method to randomly select sample sets and construct a decision tree; S32: Randomly select a feature subset for the decision tree node and perform optimal splitting; S33: Calculate the importance of the stability evaluation features of each railway slope dangerous rock using the OOB error. The calculation formula is: ； Where, For the The importance of features, For the Out-of-bag errors, To disrupt the Out-of-bag error after features, is the number of trees in the random forest.

3. The intelligent prediction method for railway dangerous rock stability according to claim 2 is characterized in that: In step S4, the global optimization of the hyperparameters of the SVM model using a genetic algorithm includes the following steps: S41: parameter initialization, including population size, crossover probability, mutation probability and number of iterations; S42: Use the fitness function Fitness to evaluate the quality of individuals. The fitness calculation formula is: Where, For individuals The fitness of For individuals The mean square error of the corresponding SVM model, It is a small positive number to avoid division by zero errors; S43: Perform a selection operation in a roulette-wheel manner to determine the probability of inheriting the next generation of individuals. The calculation formula is: , where For individuals The probability of being selected, pop is the population size; S44: Perform crossover and mutation operations, where: , where is the original value of a gene in the current chromosome before mutation, is the new value of a gene in the current chromosome after mutation, is a random number uniformly distributed in the interval [0,1] to ensure the randomness of the mutation operation. 、 are the upper and lower limits of the parameter values respectively; S45: When the convergence condition is met or the maximum number of iterations is reached, the iteration is stopped and the optimal parameters are output, including the penalty factor C and the kernel function width γ.

4. The intelligent prediction method for railway dangerous rock stability according to claim 3 is characterized in that: In step S5, the railway slope dangerous rock stability prediction model RF-GA-SVM is constructed based on the SVM model after hyperparameter optimization, including the following steps: The SVM model is defined using the RBF kernel function as: ； Where, is the input sample, For the input sample The predicted output value of is the number of training samples, is the RBF kernel function, is the sample feature vector, 、 are all Lagrange multipliers, express The weight in the decision function, In the standard classification SVM, it is always zero. is the bias term; The calculation formula of RBF kernel function is: , where is the kernel function parameter; The SVM model determines parameters by optimizing the following objective function: , where is the hyperplane normal vector, is the bias term, is the penalty factor, is the slack variable, is the number of training samples, and Constraints must be met: ,in, For the The label of each sample is +1 or -1, representing the positive and negative classes respectively.

5. The intelligent prediction method for railway dangerous rock stability according to claim 1 is characterized in that: The determination of the railway slope dangerous rock stability evaluation characteristics at least takes into account the slope dangerous rock conditions, rock mass structure characteristics, and hydrogeological conditions.

6. The intelligent prediction method for railway dangerous rock stability according to claim 1 is characterized in that: In step S2, preprocessing the initial data set includes at least: filling in missing values, removing outliers and data standardization.

Images