Landslide prediction method based on improved random forest under underground engineering scene

Abstract

The underground engineering scene based on improved random forest landslide prediction method belongs to the technical field of underground engineering, in order to solve the problem that the existing random forest cannot guarantee good precision while having good stability when predicting landslide. The present application first collects data in real underground engineering scene and carries out cleaning, trains CART using training set, applies each CART to predict on multiple test sets to obtain average classification accuracy, and sorts in descending order; the vector inner product method is used to calculate and save the inner product value between decision trees, the decision trees with vector inner product less than or equal to the inner product threshold are retained, otherwise they are marked as deletable; the decision trees marked as deletable are deleted in turn from low to high according to the classification accuracy, until the number of remaining decision trees is N, the decision result of the improved random forest is determined by combining the majority voting rule; the obtained random forest classifier is used for landslide prediction.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of underground engineering, and particularly relates to a landslide prediction method in an underground engineering scene. BACKGROUND

[0002] Underground engineering construction plays a crucial role in modern urbanization and infrastructure development, especially in the construction process of tunnels, underground spaces, subways and other projects. Geological conditions and natural disasters have become one of the main factors affecting the safety and construction progress of the project. Landslide, as a typical geological disaster, poses a serious threat to the safety of underground engineering. The occurrence of landslide not only causes huge economic losses, but also may endanger life safety. Therefore, landslide prediction and prevention are particularly important in underground engineering construction. Effective landslide prediction can provide timely warning for engineering construction, help engineering personnel take appropriate protective measures, thereby reducing disaster losses and ensuring the safety and stability of underground engineering.

[0003] Traditional landslide prediction methods are mainly based on geological surveys, empirical rules, statistical analysis and other means, relying on expert knowledge and a large amount of field data to predict the probability of landslide occurrence. Early landslide prediction methods generally use geological survey data, combined with geological and meteorological factors such as slope, geological type, soil moisture, precipitation, etc., to predict landslide by establishing risk assessment models. For example, Wang et al. proposed a landslide displacement time series prediction model based on mean low-order autoregressive tensor completion (MLATC). Xue et al. used a data-driven landslide susceptibility assessment method and a physics-based landslide stability assessment method for reservoir landslide prediction. Niu et al. developed a GIS-based landslide geological disaster prediction method for the contact zone between soil and rock in southern Shaanxi, establishing a landslide geological disaster information value prediction model. Junwei et al. proposed an ILF-FFT prediction and warning model for reservoir landslide under rainfall conditions, based on the limit curvature of velocity and acceleration time history function, realizing a four-stage quantitative division method for landslide evolution.

[0004] However, traditional landslide prediction methods have certain limitations. First, these methods usually rely on expert experience and intuition, lack systematic scientific basis and data support, and are prone to subjective and biased prediction results. Second, traditional landslide prediction methods rely on local data of geological background, which is difficult to effectively handle complex multi-dimensional and multi-variable problems in underground engineering environment. More importantly, the prediction accuracy of traditional landslide prediction methods is often limited by data quality. For large-scale landslide prediction, the adaptability and generalization ability of traditional landslide prediction methods are weak, which cannot meet the high requirements of actual engineering application. Therefore, it is urgent to develop new landslide prediction methods to improve prediction accuracy, applicability and reliability.

[0005] With the development of artificial intelligence technology, machine learning-based intelligent landslide prediction methods have gradually become a research hotspot. The principle is to analyze the relationship between historical landslide events and related environmental factors (such as topography, geology, weather, etc.), build a model to predict new landslide occurrence risks. The process first collects and processes a large amount of data, extracts key features, and then uses classifiers (such as support vector machines, decision trees, random forests, etc.) to train the data, learn the rules and patterns of landslide occurrence. The trained classifier can predict the possibility of landslides by inputting the corresponding feature data when facing new geological conditions. Machine learning can handle complex nonlinear relationships, mine potential patterns and associations, and provide more accurate and dynamic landslide predictions than traditional landslide prediction methods. Many scholars have conducted related research, such as Zhang et al. proposed a new fuzzy deep learning (FuDL) model for near real-time earthquake-induced landslide spatial prediction. Feng et al. proposed a landslide displacement prediction method based on a chaotic Gaussian variant sparrow search algorithm optimized BP neural network (CG-SSA-BP). Ma et al. used machine learning algorithms to establish prediction models for similar type landslide deformation, mixed type landslide deformation, and single landslide deformation. In particular, as an integrated learning method, random forests have been widely used in landslide prediction due to their advantages in handling high-dimensional data, especially in dealing with nonlinear problems. For example, Wang et al. used Bayesian hyperparameter optimization to optimize the hyperparameters of random forests, and then selected the optimal hyperparameters for landslide susceptibility mapping. Cheng et al. used random forests to predict landslide areas.

[0006] Although intelligent landslide prediction methods have higher accuracy and adaptability compared to traditional landslide prediction methods, there are still some problems in existing research. First, most existing machine learning classifiers are validated in laboratory environments, and the data sets used are often public data sets, which lack specific adaptability to underground engineering scenarios, resulting in poor performance of these classifiers in actual applications. For example, many studies use public data sets in geological research, which have significant differences in environmental conditions from the complex situations faced in actual underground engineering. Therefore, classifiers for landslide prediction need to combine complex data from actual engineering environments to better meet the needs of engineering applications. Second, existing research mostly focuses on selecting the most advanced classifier or optimizing existing classifiers, especially in parameter tuning, to find the best combination. However, these methods ignore the theoretical depth and optimization space of the classifier, often lacking optimization from the construction principle. For example, some studies fine-tune existing models through deep learning methods or reinforcement learning, which have achieved good results on certain data sets, but due to the lack of systematic optimization, the generalization ability and stability of the classifier are poor. Therefore, classifier selection and optimization should focus on the mutual cooperation of theoretical basis and actual data, rather than relying solely on parameter tuning. Finally, many intelligent landslide prediction methods are simple in scheme design, usually selecting a classifier for experiments, but failing to conduct a thorough comparative analysis and giving the selection basis for the optimal classifier. Although many studies have made preliminary attempts on different classifiers, they usually do not conduct detailed performance comparisons and do not explicitly state the applicability of different algorithms in specific engineering environments.

[0007] In addition, existing random forests for landslide prediction measure the correlation between features uniformly. Since the same type of features may have different differences in different environments, when a random forest is established for a certain data set, it can achieve good prediction results, but when it is used for landslide prediction in different underground engineering scenarios, it may not guarantee good recognition results, i.e., it cannot guarantee good accuracy and stability at the same time. SUMMARY

[0008] To solve the problem that existing random forests for landslide prediction cannot guarantee good accuracy and stability at the same time.

[0009] A landslide prediction method based on an improved random forest in an underground engineering scenario, which first acquires geological parameter features and then uses an improved random forest model for landslide prediction.

[0010] The construction process of the improved random forest model includes the following steps:

[0011] Step 1: Collect samples by inserting the integrated sensor module connected by the parameter acquisition terminal into the slope, each sample including multiple geological parameter characteristics, and then obtaining an original sample set;

[0012] Select K sample subsets from the original sample set by a non-replacement random sampling method, K being greater than or equal to 3, as a test set for evaluating the classification performance of each CART;

[0013] Step 2: Determine the number N of CARTs to be trained and the number of features in the corresponding feature subset; use Bagging to perform (1+m)×N times of random sampling with replacement on the remaining samples of the original sample set to construct (1+m)×N training sets; on this basis, train (1+m)×M CARTs; wherein m×N CARTs are trained in a certain number of CARTs on the basis of a preset number;

[0014] Step 3: Apply each CART to predict on the K sample subsets, and express the obtained classification accuracy as Wherein, i = 1, 2,..., (1+m)×N, represents the ith CART, and j = 1, 2, 3,..., K, represents the jth sample subset;

[0015] Step 4: Calculate the average classification accuracy of the ith CART

[0016] Step 5: Arrange the CARTs in descending order according to the obtained average classification accuracy;

[0017] Step 6: Calculate and save the inner product values between the CARTs using the vector dot product method shown in formula (1); take the overall classification accuracy of the random forest as an evaluation index, and combine the grid search method to find the optimal inner product threshold t; for a pair of CARTs with an inner product value less than or equal to the inner product threshold, keep them; for a pair of CARTs with an inner product value greater than the inner product threshold, mark the CART with a lower average classification accuracy as deletable;

[0018] Sim(D i ,D j )=acos(W i ·W j ) (1)

[0019] In the formula, D i and D j represent two CARTs, W i represents the feature subset corresponding to CARTi, W j represents the feature subset corresponding to CARTj, and acos(·) represents the inverse cosine function;

[0020] W i ·W j The detailed calculation formula is as follows:

[0021]

[0022] In the formula, W im represents the mth feature in the feature subset W i , W jn represents the mth feature in the feature subset W j ; I is an indicator function, only when W im =W jn , I(W im =W jn )=1, otherwise I(W im =W jn )=0; W im =W jn indicates that the same feature is selected in different feature subsets;

[0023] Step 7: all the CARTs marked as deletable are arranged in ascending order according to the average classification accuracy obtained, and are deleted in turn until the number of the remaining CARTs is N; that is, the number of the remaining CARTs reaches the preset number;

[0024] Step 8: the N CARTs reserved are used to construct a random forest; combined with the majority voting rule, the decision result of the improved random forest is determined.

[0025] Further, the geological parameters include: inclination angle, surface displacement, horizontal displacement, vertical displacement, water pressure, soil deformation, soil stress, soil humidity, soil temperature, shear strength and electrical conductivity.

[0026] Further, in the Step 7, if the number of the remaining CARTs that are not marked after all the CARTs marked as deletable are deleted is still greater than N, the CARTs are deleted in turn according to the sorting of the CARTs given in Step 5 until the number of the remaining CARTs is N.

[0027] Further, in Step 1, the original sample set is obtained by data cleaning on the data set corresponding to the collected samples.

[0028] Further, the K=5.

[0029] Beneficial effects:

[0030] The improved random forest based on multi-decision tree correlation measurement can improve the accuracy and stability of landslide prediction. Meanwhile, the overall and local classification performance of different classifiers and neural networks is comprehensively evaluated to provide the selection basis of the optimal classifier, thereby ensuring the final classification effect. In addition, the data in the real underground engineering application scene is collected and cleaned to construct a high-quality and high-reliability data set, thereby ensuring the prediction effectiveness and accuracy of the final random forest. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 The flow chart of the improved random forest is constructed.

[0032] Figure 2 The Precision graph obtained for the six classifiers.

[0033] Figure 3 The Recall graph obtained for the six classifiers.

[0034] Figure 4 The F1-score graph obtained for the six classifiers.

[0035] Figure 5 The relationship graph between the value of the inner product threshold and the classification accuracy obtained by the random forest.

[0036] Figure 6 The Precision graph obtained for the three improved random forests.

[0037] Figure 7 The Recall graph obtained for the three improved random forests.

[0038] Figure 8 The F1-score graph obtained for the three improved random forests.

[0039] Figure 9 The Precision graph obtained for the four improved random forests.

[0040] Figure 10 The Recall graph obtained for the four improved random forests.

[0041] Figure 11 The F1-score graph obtained for the four improved random forests. DETAILED DESCRIPTION

[0042] Underground engineering construction is crucial in modern urbanization and infrastructure development. Landslide, as a typical geological disaster, poses a serious threat to the safety of underground engineering, and landslide prediction has become an important means to ensure engineering stability. Traditional landslide prediction methods rely on expert experience and geological surveys, and although they have achieved certain results in specific scenarios, they often lack systematic data support and scientific basis, resulting in subjective and inaccurate predictions. In addition, traditional landslide prediction methods are difficult to adapt to the complexity and multidimensionality of underground engineering environments, and have poor adaptability and generalization ability when dealing with large-scale landslide prediction. Therefore, more and more research has turned to intelligent landslide prediction methods based on machine learning, which analyze the relationship between landslide historical data and geological, topographic, and meteorological factors to establish data-driven models to predict landslide likelihood. However, existing intelligent landslide prediction methods mostly use public datasets for validation, lacking adaptability to underground engineering scenarios, limiting their practical application effectiveness.

[0043] The present invention aims to address the limitations of existing traditional and intelligent landslide prediction methods in underground engineering scenarios. First, in a real underground engineering scenario, special equipment is used to collect key parameters of landslide occurrence on site, and the collected raw data is used to build a preliminary dataset. Subsequently, data cleaning is performed on the dataset, including missing value and outlier processing, to ensure data quality and reliability, laying a solid foundation for subsequent modeling. Second, various latest or classic classifiers and neural networks suitable for data classification are trained on the dataset, and their overall classification performance on the dataset and their local classification performance on different labeled data are analyzed in detail. Through comprehensive evaluation, the random forest is selected as the optimal classifier, and the scientific basis for selection and its adaptability in specific engineering environments are elaborated. Finally, based on a deep understanding of the principles of random forest construction, an improved random forest method based on correlation measurement of multiple decision trees is proposed, and a detailed comparative study is conducted with the classic random forest and other improved versions. The results show that the improved random forest performs better in landslide prediction, further improving the prediction accuracy and stability, effectively addressing the limitations of traditional and intelligent landslide prediction methods in underground engineering scenarios. The present invention effectively addresses the shortcomings of existing landslide prediction methods, not only providing a new solution for landslide prediction in underground engineering scenarios, but also providing important reference and reference for the development of theory and practice in the field of landslide prediction. DETAILED DESCRIPTION

[0045] The embodiment is a landslide prediction method based on an improved random forest in an underground engineering scene. First, an improved random forest model is constructed, and then the improved random forest model is used for landslide prediction. The random forest is a combination classifier based on ensemble learning, which is composed of multiple CART decision trees. The process of constructing the improved random forest model is as shown in Figure 1

[0046] Step 1: The integrated sensor module connected by the parameter acquisition terminal is inserted into the slope to collect samples. Each sample includes multiple geological parameter characteristics, including: inclination angle, surface displacement, horizontal displacement, vertical displacement, water pressure, soil deformation, soil stress, soil humidity, soil temperature, shear strength, and electrical conductivity. Subsequently, the original sample set is obtained by data cleaning on the data set.

[0047] Five sample subsets are selected from the original sample set by non-replacement random sampling as test sets for evaluating the classification performance of each CART.

[0048] It should be noted that the five sample subsets contain the same number of samples, but the specific samples contained are different, and the labels of these samples are known.

[0049] Step 2: Determine the number N of CARTs to be trained and the number of features in the corresponding feature subset. Use Bagging to perform (1+m)×N times of random sampling with replacement on the remaining samples of the original sample set to construct (1+m)×N training sets. On this basis, train (1+m)×N CARTs. Among them, m×N CARTs are trained a certain number of CARTs based on the preset number.

[0050] It should be noted that m is called a weighting coefficient, and the value is usually 0.1 or 0.15.

[0051] Step 3: Apply each CART to predict on the five sample subsets, and express the obtained classification accuracy as Where i=1, 2,..., (1+m)×N, represents the ith CART, and j=1, 2, 3, 4, 5, represents the jth sample subset.

[0052] Step 4: Calculate the average classification accuracy of the ith CART, and the formula is:

[0053] Step 5: Arrange the CARTs in descending order according to the average classification accuracy obtained.

[0054] ​Step 6: Calculate and save the inner product value between CARTs using the vector dot product method shown in formula (1). Take the overall classification accuracy obtained by random forest as the evaluation index, and combine the grid search method to find the optimal inner product threshold t. In this way, for those pairs of CARTs whose inner product values are less than or equal to the inner product threshold, no processing is performed. For those pairs of CARTs whose inner product values are greater than the inner product threshold, mark the CART with lower average classification accuracy as deletable.

[0055] Sim(D i ,D j )=acos(W i ·W j ) (1) i ,D j )=acos(W i ·W j ) (1)

[0056] In the formula, D i and D j represent two CARTs, W i represents the feature subset corresponding to CARTi, W j represents the feature subset corresponding to CARTj, and the detailed calculation formula of W i ·W j is as follows:

[0057]

[0058] In the formula, W im represents the mth feature in the feature subset W i , W jn represents the nth feature in the feature subset W j , and I is an indicator function. I(W im =W jn ) = 1 only when W im =W jn , otherwise I(W im =W jn ) = 0. W im =W jn indicates that the same feature is selected in different feature subsets.

[0059] As can be seen from formula (2), the denominator part is to calculate the sum of the values between the feature subset W i and all features in the feature subset W j , and the numerator part is to calculate the value between the feature subset W i and the feature subset W jThe sum of the values between the same features. It can be seen that the key to measuring the correlation between two CARTs is the numerator part of formula (2). It is worth noting that in the present application, the values between the corresponding feature subsets of the CARTs, i.e. the features, are actually used to measure the correlation between the two. Through the study of landslides and the corresponding geological parameters, the present application determines to use the dot product method of formula (1) to evaluate the correlation between the CARTs by evaluating the role of the same geological parameters in different feature subsets to assist the judgment of the consistency. Through subsequent verification, the present application can improve the accuracy and stability of landslide prediction, and can have good stability while ensuring good accuracy.

[0060] Step 7: Arrange all the CARTs marked as deletable in ascending order according to the average classification accuracy obtained, and delete them in turn until the number of remaining CARTs is N. That is, the number of remaining CARTs reaches the preset number. It should be noted that if the CARTs marked as deletable are all deleted, and the number of those remaining CARTs that are not marked is still greater than N, then delete them in turn according to the ranking of the CARTs given in Step 5 until the number of remaining CARTs is N.

[0061] Step 8: Use the remaining N CARTs to build a random forest. Combine the majority voting rule to determine the decision result of the improved random forest.

[0062] Embodiment

[0063] The specific implementation process of the present embodiment is as follows:

[0064] 1. Complete data set:

[0065] Firstly, the collection terminal is placed on multiple slopes in the underground engineering scene, and the integrated sensor module connected to the parameter collection terminal is inserted into the slope to collect geological parameters, including: inclination angle, surface displacement, horizontal displacement, vertical displacement, water pressure, soil deformation, soil stress, soil humidity, soil temperature, shear strength and electrical conductivity, a total of 11 geological parameters, referred to as 11 features. In this way, the parameter collection terminal will obtain the numerical values of 11 geological parameters each time it is collected during operation, forming a data, and multiple collections will eventually obtain multiple data. According to the actual needs of landslide prediction in the underground engineering scene, these data are divided into four categories, including: normal, small-scale soil erosion, soil erosion and landslide, quantified as "0", "1", "2" and "3" four labels, and the corresponding labels are added to these data. Taking the underground engineering of a province in southern China as an example, a large amount of data was collected using the parameter collection terminal from January 2022 to October 2023, and the above labels were added to the data according to the actual situation to construct a preliminary data set, the detailed description of which is shown in Table 1. It is worth noting that before constructing the preliminary data set, the original large amount of data collected was artificially balanced in quantity to ensure that the number of data of the four labels is approximately equal. Specifically, the number of data with label "3" is taken as the reference number, and approximately equal number of data of the other three labels are selected respectively.

[0066] Table 1 Detailed description of the preliminary data set

[0067]

[0068] After statistics, the preliminary data set contains 30389 data. According to the 3σ principle of normal distribution, the preliminary data set is detected for abnormal values. Specifically, taking 11 features as the basic unit, the 30389 values belonging to the same feature are detected each time, and the values exceeding three standard deviations are regarded as abnormal values. In this way, those data with abnormal values are called abnormal data. Similarly, the preliminary data set is detected for missing values. Specifically, the values with "NULL" and special symbols such as "?" are regarded as missing values. In this way, those data with missing values are called missing data. On this basis, the abnormal data and the missing data are collectively referred to as incomplete data. It is worth noting that there may be some data in the incomplete data that contains both abnormal values and missing values. After statistics, there are 43 incomplete data in the data with label "0", 51 incomplete data in the data with label "1", 45 incomplete data in the data with label "2", and 56 incomplete data in the data with label "3", a total of 195 incomplete data. Sun et al. showed that for a data set, if the proportion of abnormal data does not exceed 2%, or the proportion of missing data does not exceed 3%, then these abnormal data or missing data can be directly discarded, because they do not pose a threat to the completeness and diversity of the data set.

[0069] After calculation, the proportion of incomplete data with label "0" in the data with label "0" is about 0.57%, the proportion of incomplete data with label "1" in the data with label "1" is about 0.67, the proportion of incomplete data with label "2" in the data with label "2" is about 0.59, the proportion of incomplete data with label "3" in the data with label "3" is about 0.74%, and the proportion of all incomplete data in the data set is about 0.64%. It can be seen that the proportion of incomplete data does not exceed 1% in each label data or in the entire data set, not to mention the proportion of abnormal data or missing data. Therefore, the incomplete data in the preliminary data set is directly removed, and a high-quality complete data set under the underground engineering scene is obtained for subsequent experiments, and the detailed description is shown in Table 2.

[0070] Table 2 Detailed description of complete data set

[0071]

[0072] 2、Optimal classifier:

[0073] Classifiers in machine learning are generally divided into three types, classifiers based on mathematical theory, classifiers based on tree models and deep learning models. In the classifiers based on mathematical theory, k-nearest neighbor and support vector machine are the most commonly used. In the classifiers based on tree models, decision tree and random forest are the most commonly used. In the deep learning models, multilayer perceptron and one-dimensional convolutional neural network are the most commonly used for data classification. Therefore, for the complete data set shown in Table 2, it is divided into training set and test set according to the ratio of 7:3. The above six classifiers are trained on the training set respectively, and the parameter configuration when they achieve the highest classification accuracy is obtained, as shown in Table 3, and their classification performance is tested again on the test set, and the classification accuracy, Precision, Recall and F1-score they achieve are counted. Table 4 gives the classification accuracy of the six classifiers on the test set. Figures 2 to 4 The Precision, Recall and F1-score of the six classifiers on the test set are given.

[0074] Table 3 Optimal parameter configuration of six classifiers

[0075]

[0076] Table 4 Classification accuracy of six classifiers

[0077]

[0078] As can be seen from Table 4, the classification accuracy of random forest is the highest, which is 89.84%, which is higher than the classification accuracy of k-nearest neighbor, support vector machine, decision tree, multilayer perceptron and one-dimensional convolutional neural network by 4.69%, 1.81%, 3.43%, 10.92% and 2.07% respectively, which has obvious advantages. Among them, the support vector machine achieves the second highest classification accuracy, which has obvious performance advantage compared with the other four classifiers, which proves that it can construct the boundary surface of high-dimensional feature space to effectively realize data classification under the condition of introducing kernel function, but compared with random forest based on ensemble learning, there is still a certain gap. One-dimensional convolutional neural network also achieves satisfactory classification accuracy, which proves its stability as a classic neural network for data classification. Among them, the multilayer perceptron achieves the lowest classification accuracy, and its classification performance is greatly affected by the initial weight, which is easy to fall into local optimal solution. From Figures 2 to 4As can be seen, Random Forest achieves the highest Precision, Recall, and F1-score on the four-label data, demonstrating a significant advantage over other classifiers. This proves that it achieves the most significant overall classification performance on the dataset. Furthermore, Random Forest also exhibits the most balanced and stable Precision, Recall, and F1-score across the four-label data, with minimal differences between them. This proves that it achieves the most significant and stable local classification performance on the data for each label.

[0079] 3. Improve Random Forest:

[0080] The improved random forest method based on the multi-decision tree correlation metric of this invention improves both the random forest with default parameter configuration and the random forest with optimized parameters (optimal parameter configuration in the optimal classifier), resulting in two improved random forests, referred to as the default improved random forest and the optimized improved random forest, respectively. The inner product thresholds for the two improved random forests are set to 21 and 25, respectively, as explained below. As mentioned earlier, the inner product threshold needs to be determined to judge the correlation between CARTs. Specifically, the overall classification accuracy achieved by the random forest is used as the evaluation index, and a grid search method is used to find the optimal inner product threshold. Multiple CARTs are trained on the training set, and the inner product values ​​between CARTs are calculated. Based on this, and empirically, the range of the inner product threshold t is set to 5 to 29. Thus, when the inner product threshold t is sequentially selected between 5 and 29, the pair of CARTs with lower average classification accuracy above the inner product threshold t is deleted, and the remaining CARTs are used to construct the random forest, and its classification accuracy on the test set is calculated. Therefore, using the grid search method, the range of the search inner product threshold t is set to 5 to 29. Finally, 25 random forests were constructed, resulting in 25 classification accuracies. The optimal inner product threshold t can then be determined by comparing which random forest achieves the highest average classification accuracy. Figure 5 The relationship between the inner product threshold value and the classification accuracy achieved by random forest is presented.

[0081] Depend on Figure 5 As can be seen, the default improved random forest achieves the highest classification accuracy when the inner product threshold is 21. The optimized improved random forest achieves the highest classification accuracy when the inner product threshold is 25. Thereafter, as the inner product threshold continues to increase, the average classification accuracy of both improved random forests decreases slightly before stabilizing. Therefore, in this invention, the inner product threshold for the default improved random forest is set to 21, and the inner product threshold for the optimized improved random forest is set to 25.

[0082] On this basis, the classification accuracy, Precision, Recall and F1-score of the two improved random forests are counted. The classification accuracy of the default improved random forest and the optimized improved random forest is 91.68% and 93.05%, respectively. Compared with the classification accuracy of 89.84% of the random forest after parameter optimization, it is increased by 1.84% and 3.21%, respectively. This preliminarily shows the feasibility, practicability and superiority of the improved random forest method proposed in the application. The advantage of the classification accuracy of the default improved random forest compared with the random forest after parameter optimization fully shows that only fine-tuning the parameters of the classifier cannot improve its generalization ability and stability, but systematic optimization from the construction principle can effectively solve this problem. This fully shows the necessity of carrying out the application. The most significant classification accuracy of the optimized improved random forest fully shows that simultaneous parameter fine-tuning and systematic optimization can simultaneously improve the classification performance of the original classifier, which is a worthy integrated optimization scheme. In addition, Figures 6 to 8 The Precision, Recall and F1-score of the three improved random forests (including the random forest after parameter optimization) on the test set are given.

[0083] From Figures 6 to 8 It can be seen that the optimized improved random forest has the highest Precision, Recall and F1-score on the data of the four labels, which is obviously superior to the default improved random forest and the parameter optimization random forest. This proves that the overall classification performance of the optimized improved random forest on the data set is the most significant. At the same time, the Precision, Recall and F1-score of the default improved random forest are higher than those of the parameter optimization random forest. This effectively shows the feasibility and superiority of the improved random forest method proposed in the application. Moreover, the Precision, Recall and F1-score of the optimized improved random forest on the data of the four labels are also the most balanced and stable, with the smallest difference between each other. This proves that the local classification performance of the optimized improved random forest on the data of each label is the most significant and stable. At the same time, the difference between the Precision, Recall and F1-score of the default improved random forest is smaller than that of the parameter optimization random forest. This effectively shows the stability of the improved random forest method proposed in the application. In summary, the improved random forest method proposed in the application can simultaneously improve the overall classification performance and local classification performance of the random forest, which is obviously superior to the existing parameter optimization method.

[0084] The improved random forests in the latest literature are used for comparison in landslide prediction, including the Bayesian hyper-parameter optimization random forest (BH-RF) proposed by Wang et al., the Bayesian optimization random forest combined Kalman filter (BORF-KF) proposed by Zhang et al., and the cascade forest random forest framework (CF-RF) proposed by Chen et al. The above three improved random forests are trained on the training set, and their classification performance is tested on the test set. The classification accuracy, Precision, Recall and F1-score achieved by them are counted for comparison with the optimized improved random forest. Table 5 gives the classification accuracy achieved by the four improved random forests on the test set. Figures 9 to 11 Table 6 gives the Precision, Recall and F1-score achieved by the four improved random forests on the test set.

[0085] Table 5 Classification accuracy achieved by the four improved random forests

[0086]

[0087] As can be seen from Table 5, the optimized improved random forest constructed by the present application still achieves the highest classification accuracy of 93.05%, which is 2.69%, 2.77% and 3.13% higher than the classification accuracy of 90.36%, 90.28% and 89.92% achieved by BH-RF, BORF-KF and CF-RF, respectively, and has obvious advantages. It is worth noting that the classification accuracy achieved by the default improved random forest constructed is 91.68%, which is also higher than the classification accuracy achieved by BH-RF, BORF-KF and CF-RF. This fully demonstrates the superiority of the improved random forest method proposed by the present application, which achieves the best overall classification performance on the data set. In fact, BH-RF is also an improved random forest method for parameter optimization, and BORF-KF and CF-RF are improved random forest methods combining random forest with other classifiers. As described in the introduction, these methods focus on optimizing the existing classifiers, especially a large number of explorations on parameter tuning, or using reinforcement learning to fine-tune the existing model, and none of them can start from the construction principle of random forest to carry out systematic optimization research. This fully proves the correctness and necessity of carrying out the present application.

[0088] From Figures 9 to 11As can be seen, the optimized and improved random forest constructed by the present application achieves the highest Precision, Recall and F1-score, which is obviously higher than the Precision, Recall and F1-score achieved by BH-RF, BORF-KF and CF-RF. Moreover, the Precision, Recall and F1-score achieved by the optimized and improved random forest on the data of the four labels are also the most balanced and stable, and the difference between them is the smallest. This again shows that the optimized and improved random forest constructed by the present application achieves the most significant and stable local classification performance on the data of each label, and demonstrates its local classification performance advantage compared with BH-RF, BORF-KF and CF-RF. In summary, compared with the existing improved random forest for landslide prediction, the overall classification performance and local classification performance of the optimized and improved random forest constructed by the present application are both better. The feasibility and correctness of the improved random forest method proposed by the present application are proved, and its feasibility, practicability, stability and superiority are fully proved.

[0089] The above examples of the present application are only to illustrate the calculation model and calculation process of the present application, and are not limited to the embodiments of the present application. Based on the above description, those skilled in the art can make other different forms of changes or modifications, and it is impossible to enumerate all the embodiments here. Any obvious changes or modifications derived from the technical solutions of the present application are still within the protection scope of the present application.

Claims

1. A landslide prediction method based on improved random forest in underground engineering scene, characterized in that, Firstly, geological parameter characteristics are acquired, and then improved random forest model is used for landslide prediction; The construction process of the improved random forest model comprises the following steps: Step 1: An integrated sensor module connected through a parameter acquisition terminal is inserted into a slope to collect samples, each sample set comprises a plurality of geological parameter characteristics, and then an original sample set is obtained; K sample subsets are selected from the original sample set by a non-replacement random sampling method, K is greater than or equal to 3, and the K sample subsets are used as test sets for evaluating the classification performance of each CART; Step 2: The number N of CARTs to be trained and the number of features in the corresponding feature subset are determined; (1+m)×N times of random sampling with replacement are performed on the remaining samples of the original sample set by using Bagging to construct (1+m)×N training sets; on this basis, (1+m)×N CARTs are trained; wherein, m×N CARTs are trained in a certain number of CARTs on the basis of a preset number; Step 3: Apply each CART to make prediction on K subsets of samples, and express the obtained classification accuracy as where i = 1, 2,..., (1 + m) x N, represents the i-th CART, and j = 1, 2, 3,..., K, represents the j-th subset of samples. Step 4: Calculate the average classification accuracy achieved by the ith CART Step 5: The CARTs are arranged in descending order according to the obtained average classification accuracy; Step 6: The vector dot product method shown in formula (1) is used to calculate and save the inner product values between the CARTs; the overall classification accuracy obtained by the random forest is used as an evaluation index, and a grid search method is used to find an optimal inner product threshold t; for a pair of CARTs with an inner product value less than or equal to the inner product threshold, the pair of CARTs are retained; for a pair of CARTs with an inner product value greater than the inner product threshold, the CART with a lower average classification accuracy is marked as deletable; Sim(D i ,D j ) = aco s(W i · W j ) (1) where D i and D j respectively represent two CARTs, W i represents the feature subset corresponding to CARTi, W j represents the feature subset corresponding to CARTj, and acos(·) represents the inverse cosine function. W i • W j The detailed calculation formula is as follows: where W im represents the mth feature in the feature subset W i ; W jn represents the nth feature in the feature subset W j ; I is an indicator function, I(W im = W jn ) = 1 only when W im = W jn , otherwise I(W im = W jn ) = 0; W im = W jn represents that the same feature is selected in different feature subsets. Step 7: All the CARTs marked as deletable are arranged in ascending order according to the obtained average classification accuracy, and are sequentially deleted until the number of the remaining CARTs is N; that is, the number of the remaining CARTs reaches the preset number; Step 8: The N retained CARTs are used to construct a random forest; and a majority voting rule is used to determine the decision result of the improved random forest.

2. The landslide prediction method based on improved random forest in underground engineering scene according to claim 1, characterized in that, The geological parameters comprise: inclination angle, ground displacement, horizontal displacement, vertical displacement, water pressure, soil deformation, soil stress, soil humidity, soil temperature, shear strength and electrical conductivity.

3. The landslide prediction method based on improved random forest in underground engineering scene according to claim 1 or 2, characterized in that, In Step 7, if all the CARTs marked as deletable are deleted, and the number of the remaining CARTs that are not marked is greater than N, then the CARTs are sequentially deleted according to the sorting of the CARTs given in Step 5 until the number of the remaining CARTs is N.

4. The landslide prediction method based on improved random forest in underground engineering scene according to claim 3, characterized in that, In Step 1, data cleaning is performed on the data set corresponding to the collected samples to obtain the original sample set.

5. The landslide prediction method based on improved random forest in underground engineering scene according to claim 3, characterized in that, The K is 5.

Images