A method for predicting displacement rate of step-type landslide based on LSO-RF model

Abstract

The present invention relates to a method for predicting the displacement rate of a step-type landslide based on an LSO-RF model, comprising the following steps: Step S1: collecting and processing landslide-related data to obtain the landslide displacement rate and related factors in daily units; Step S2: comprehensively considering the degree of correlation between the landslide displacement rate and various factors based on the results of the Spearman rank correlation coefficient method and the grey correlation analysis method, and doubly selecting the input features of the prediction model; Step S3: constructing a data set using the input features obtained in Step S2 and the historical data of the landslide displacement rate; Step S4: constructing an LSO-RF model, training it based on the training set data, obtaining the trained LSO-RF model, and predicting the landslide displacement rate based on the trained LSO-RF model. The present invention can achieve accurate prediction of the displacement rate of a step-type landslide.

Description

Technical Field

[0001] The invention belongs to the field of landslide monitoring and prediction, and in particular relates to a step-type landslide displacement rate prediction method based on an LSO-RF model. Background Art

[0002] Current research on landslide deformation trend prediction focuses primarily on predicting the cumulative displacement of a landslide, while relatively little research has focused on predicting the sliding rate. Compared to cumulative displacement, the sliding rate metric is more intuitive and can be better applied to landslide early warning by categorizing the displacement rate into different levels.

[0003] A step-type landslide is a type of landslide that exhibits alternating changes between "step sections" and "stable sections" in its cumulative displacement. It is widely found in the Three Gorges Reservoir area and the hilly and mountainous areas along the southeastern coast. The sliding rate of this type of landslide often presents the following characteristics: (1) When the landslide is in the "stable section", its displacement rate is very small, and the landslide only undergoes relatively small movements; (2) When the landslide is in the "step section", its displacement rate suddenly rises from the above-mentioned smaller value to a larger value, and then changes at a certain rate range; (3) Affected by factors such as rainfall, in general, there are more "stable section data" than "step section data" in step-type landslides. The above characteristics cause most artificial intelligence models to overfit when training the displacement rate of step-type landslides, which in turn leads to poor prediction of the displacement rate of this type of landslide and hysteresis at the peak. Therefore, an artificial intelligence model that can be applied to the characteristics of the displacement rate of step-type landslides is needed to achieve accurate prediction of the displacement rate of step-type landslides. Summary of the Invention

[0004] In view of this, the object of the present invention is to provide a method for predicting the displacement rate of a step-type landslide based on the LSO-RF model, so as to achieve accurate prediction of the displacement rate of a step-type landslide.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for predicting the displacement rate of a step-type landslide based on the LSO-RF model comprises the following steps:

[0007] Step S1: collecting and processing landslide-related data to obtain data on factors such as landslide displacement rate, rainfall, and pore water pressure in daily units;

[0008] Step S2: Based on the results of the Spearman rank correlation coefficient method and the grey correlation analysis method, the correlation degree between the landslide displacement rate and various factors is comprehensively considered, and the input features of the prediction model are double selected;

[0009] Step S3: constructing a data set using the input features and landslide displacement rate historical data obtained in step S2;

[0010] Step S4: constructing an LSO-RF model, training it based on the training set data to obtain a trained LSO-RF model, and predicting the landslide displacement rate based on the trained LSO-RF model.

[0011] Furthermore, step S1 is specifically as follows: in units of hours, the cumulative displacement, pore water pressure, and rainfall of the landslide are obtained through a field monitoring station; the cumulative displacement, pore water pressure, and rainfall data are processed and converted into the landslide displacement rate, daily average pore water pressure, and daily rainfall in units of days; and in units of days, the cloud cover, humidity, effective soil water holding capacity, and surface runoff in the landslide area are obtained through meteorological station data.

[0012] Furthermore, the step S2 is specifically as follows:

[0013] S21: Consider the landslide displacement rate as the dependent variable Y, and the landslide daily average pore water pressure, daily rainfall, cloud cover, humidity, soil effective water holding capacity, and surface runoff as the independent variables X, and explore the Spearman rank correlation coefficient between X and Y respectively;

[0014] S22: The landslide displacement rate time series is regarded as the reference time series, and the landslide daily average pore water pressure, daily rainfall, cloud cover, humidity, soil effective water holding capacity, and surface runoff time series are regarded as the comparison time series. After averaging the reference time series and the comparison time series, the grey correlation degree between the reference time series and each comparison time series is calculated.

[0015] S23: Comprehensively compare the analysis results of Spearman rank correlation coefficient and grey relational degree, and select factors that meet the preset requirements in both analyses as the input features of the model.

[0016] Furthermore, the calculation steps of the Spearman rank correlation coefficient are as follows:

[0017] (1) First, sort the data of the two variables (X, Y) separately, and then record the position after sorting; at this time, the position of a certain data i is X i or Y i rank;

[0018] (2) Then the Spearman correlation coefficient r s Calculated as follows:

[0019]

[0020] where d i For X i With Yi The rank difference between them; n is the number of samples in the sequence.

[0021] Furthermore, the calculation steps of the grey relational degree are as follows:

[0022] (1) The time series of landslide displacement rate was determined as the reference series, and the time series of landslide daily average pore water pressure, daily rainfall, cloud cover, humidity, effective soil water holding capacity, and surface runoff were determined as the comparison series;

[0023] (2) For the reference sequence {x0 } and compare the sequence {x i}, i=1,2,…,5 are dimensionless processed by averaging. The averaging formula is as follows:

[0024]

[0025] Among them, x i (k) is the kth data in a time series; is the average value of a time series;

[0026] (3) Calculate the grey correlation coefficient between the reference sequence and the comparison sequence. The calculation formula is as follows:

[0027]

[0028] Among them, ξ i (k) is the comparison sequence {x i} is the grey correlation coefficient with the reference sequence {x0} at the kth data; ρ is the resolution coefficient;

[0029] (4) Calculate the grey correlation value between the reference sequence and each comparison sequence according to the following formula:

[0030]

[0031] Furthermore, the LSO-RF model is constructed as follows:

[0032] S41: Population initialization, including initializing the number of lions in the lion group and the position of the lions;

[0033] S42: Input the RF random forest model and select the parameters to be optimized;

[0034] S43: Calculate the number of lion kings, lionesses, and lion cubs in the lion group according to the formula, and set the initial group optimal position as the lion king's position, and the current position of each lion as the individual historical optimal position;

[0035] S44: Set the fitness function of the LSO lion group optimization algorithm to the MAE, RMSE, and R of the RF random forest model prediction results.2 The average value of

[0036] S45: Update the lion king's position and calculate the fitness value;

[0037] S46: Update the position of the lioness and the position of the cubs;

[0038] S47: Recalculate the fitness value and update the global optimal position and the historical optimal position;

[0039] S48: Determine whether the number of iterations is the maximum number of iterations. If not, repeat steps S45 to S47. If it is the maximum number of iterations, end the loop and output the value of the parameter to be optimized.

[0040] S49: Based on the obtained parameters, an LSO-optimized RF random forest model is established.

[0041] Furthermore, step S4 also uses the conversion of supervised learning data to convert the time series containing different features into a time series with specified input step size and output step size.

[0042] Furthermore, when training the model in step S4, an expansion window method is used.

[0043] (1) First, the training set and the test set are divided into two sets according to a preset ratio. The model is trained on the training set and then predicts the data starting from the starting position in the test set.

[0044] (2) After the prediction of the above part of the data is completed, the true values ​​of these data are included in the original training set to form a new training set, and then the next part of the data is predicted.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The present invention solves the inaccuracy and hysteresis caused by overfitting and other reasons in most artificial intelligence models when predicting the displacement rate of step-type landslides, and can achieve more accurate predictions using fewer data sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flow chart of the method of the present invention;

[0048] Figure 2 This is a comparison diagram of the prediction results of the present invention using the Spearman rank correlation coefficient method and the grey relational analysis method for dual input feature selection and the prediction results of the present invention using only the Spearman rank correlation coefficient method or the grey relational analysis method for input feature selection;

[0049] Figure 3 A schematic diagram of the structure of the LSO-RF model of the present invention is established;

[0050] Figure 4 This is a comparison chart of the prediction results of the RF random forest model in the present invention and other models;

[0051] Figure 5 This is a comparison chart of the prediction results of the LSO-RF model and the WOA-RF model of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0053] Please refer to Figure 1 The present invention provides a method for predicting the displacement rate of a step-type landslide based on the LSO-RF model, which mainly includes the following steps:

[0054] S1: Collect and process landslide-related data through field monitoring stations and meteorological stations to obtain time series data including daily landslide displacement rate, daily average pore water pressure, daily rainfall, cloud cover, humidity, effective soil water holding capacity, and surface runoff;

[0055] S2: In the acquired data, the correlation between the landslide displacement rate and various factors is comprehensively considered based on the results of the Spearman rank correlation coefficient method and the grey relational analysis method, thus realizing the input features of the dual selection prediction model;

[0056] In this embodiment, step S2 includes the following steps:

[0057] S21: Consider the landslide displacement rate as the dependent variable Y, and the landslide daily average pore water pressure, daily rainfall, cloud cover, humidity, soil effective water holding capacity, and surface runoff as the independent variables X, and explore the Spearman rank correlation coefficient between X and Y respectively;

[0058] S22: The landslide displacement rate time series is considered as the reference time series, and the landslide daily average pore water pressure, daily rainfall, cloud cover, humidity, soil effective water holding capacity, and surface runoff time series are considered as the comparison time series. After averaging the reference time series and the comparison time series, the grey correlation degree between the reference time series and each comparison time series is calculated.

[0059] S23: By comprehensively comparing the results of the Spearman rank correlation coefficient and the grey correlation coefficient, factors that rank high in both analyses are selected as input features for the model, thereby achieving dual selection of input features for the prediction model. For example, using this method, we selected daily rainfall and pore water pressure.

[0060] Reference Figure 2 , Figure 2The results show the prediction results of using the Spearman rank correlation coefficient method alone to select input features in the RF model (input features are: soil effective water holding capacity, daily average pore water pressure), the prediction results of using the grey correlation analysis method alone to select input features (input features are: daily rainfall, surface runoff), and the prediction results of the dual selection of model input features based on the Spearman rank correlation coefficient method and the grey correlation analysis method (the method of the present invention) (input features are: daily rainfall, daily average pore water pressure). 2 As an evaluation index of the prediction effect, the results show that the method of dual selection of model input features based on the Spearman rank correlation coefficient method and the grey relational analysis method in the present invention can improve the prediction accuracy of the model.

[0061] S3: The features selected by the above method and the historical data of landslide displacement rate are used as input features of the prediction model, and the landslide displacement rate is used as the output feature (label);

[0062] In this embodiment, the historical data of landslide displacement rate is the landslide displacement rate of the previous day. Therefore, in this case, the input features of the prediction model are daily rainfall, daily average pore water pressure, and the landslide displacement rate of the previous day; the output feature is: landslide displacement rate.

[0063] S4: Build the scikit-learn machine learning framework on Windows and construct the LSO-RF model, which combines supervised learning with the expanding window method. Initially, the model is trained using the first 70% of the dataset as the training set, and the last 30% of the dataset as the test set. This data is fed into the trained LSO-RF model to predict landslide displacement rates and verify the prediction results.

[0064] Specifically, in S4, supervised learning data conversion involves converting a time series containing different features into a time series with specified input and output step sizes. In this case, the input step size is 1 and the output step size is 1. Combined with the aforementioned selection of input features, the original time series is converted into a combination of daily rainfall, daily average pore water pressure, and landslide displacement rate at step t-1, and daily rainfall, daily average pore water pressure, and landslide displacement rate at step t.

[0065] In this embodiment, the extended window method is specifically as follows:

[0066] First, the training set and test set are divided into a ratio of 7:3. That is, the data from February 25, 2021 to June 9, 2021 is used as the training set, and the data from June 10, 2021 to July 25, 2021 is used as the test set. The model is then trained on the training set and then predicted on the data from June 10, 2021.

[0067] The true value of the data on June 10, 2021 is included in the original training set (the training set at this time is the data from February 25, 2021 to June 10, 2021), and then the data on June 11, 2021 is predicted, and so on.

[0068] Reference Figure 3 , Figure 3 A structural diagram of the LSO-RF model of the present invention is established; the four parameters n_estimators, min_samples_split, min_samples_leaf, and max_features in the RF model (the parameters to be optimized in this case) are optimized using the LSO lion group optimization algorithm, and the global optimal position is continuously updated in the lion group optimization algorithm to obtain the optimal parameter combination of the RF random forest model.

[0069] Specifically, the LSO-RF model establishment includes the following steps:

[0070] S41: Population initialization, including initializing the number of lions in the lion group, the position of the lions, etc.;

[0071] S42: Input the RF random forest model and select the parameters to be optimized;

[0072] S43: Calculate the number of lion kings, lionesses, and lion cubs in the lion group according to the formula, and set the initial group optimal position to the lion king position, and the current position of each lion to the individual historical optimal position. The number of lion kings is 1, and the calculation formula for the number of lionesses and lion cubs is as follows:

[0073] n1=N*β-1

[0074] n2=N*(1-β)

[0075] Where n1 is the number of lionesses; n2 is the number of cubs; β is the proportion of adult lions, which is 0.2;

[0076] S44: Set the fitness function of the LSO lion group optimization algorithm to the MAE, RMSE, and R of the RF random forest model prediction results. 2 The average value of

[0077] S45: Update the lion king's position according to the formula and calculate the fitness value; the formula for updating the lion king's position is as follows:

[0078] x i k+1 =g k (1+γ||ρ i k -g k ||)

[0079] Among them, γ is a random number that obeys the normal distribution N(0,1); gk is the optimal position of the k-th generation lion group; p i k is the historical optimal position of the i-th lion in the k-th generation;

[0080] S46: Update the position of the lioness and the position of the cubs according to the formula. The formula for updating the position of the lioness is as follows:

[0081]

[0082] Among them, p c k is the historical best position of a randomly selected partner from the kth generation lioness; α f A disturbance factor for the lioness's range of movement, designed to enhance local search capabilities;

[0083] The formula for updating the position of the lion cub is as follows:

[0084]

[0085] Among them, p m k It represents the best historical position of the kth generation when the cub follows the lioness; represents the position where the kth lion cub was driven away within the hunting range; α c is the lion cub disturbance factor; q is a randomly generated probability factor that obeys the uniform distribution U[0,1];

[0086] S47: Recalculate the fitness value and update the global optimal position (designated as the lion king position) and the historical optimal position;

[0087] S48: Determine whether the number of iterations is the maximum number of iterations. If not, repeat steps S45 to S47. If it is the maximum number of iterations, end the loop and output the values ​​of n_estimators, min_samples_split, min_samples_leaf, and max_features.

[0088] S49: Based on the obtained parameters, an LSO-optimized RF random forest model is established.

[0089] Reference Figure 4 , Figure 4 This chart compares the prediction results of the RF random forest model used in the present invention with those of the SVM, PLS, and KNN models. This demonstrates the superiority of the RF random forest model, which combines supervised learning with a sliding window method, in predicting step-type landslide displacement rates. It addresses the low prediction accuracy and hysteresis of displacement rate peaks that are common in conventional models for step-type landslide displacement rates due to overfitting, effectively demonstrating the advanced nature of the present invention.

[0090] Reference Figure 5 , a comparison of the prediction results of the LSO-RF model and the WOA-RF model in the present invention. This demonstrates that the LSO-RF model, which combines supervised learning and a sliding window approach, achieves higher prediction accuracy than the WOA-RF model. Specifically, the LSO Lion Group Optimization algorithm achieves better results in optimizing RF model parameters than the WOA Whale Optimization algorithm, effectively demonstrating the advanced nature of the present invention.

[0091] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. A method for predicting the displacement rate of step-type landslides based on the LSO-RF model, characterized in that: The following steps are involved: Step S1: collecting and processing landslide-related data to obtain landslide displacement rate, rainfall, and pore water pressure factor data in daily units; Step S2: Based on the results of the Spearman rank correlation coefficient method and the grey correlation analysis method, the correlation degree between the landslide displacement rate and various factors is comprehensively considered, and the input features of the prediction model are double selected; Step S3: constructing a data set using the input features and landslide displacement rate historical data obtained in step S2; Step S4: constructing an LSO-RF model, training it based on the training set to obtain a trained LSO-RF model, and predicting the landslide displacement rate based on the trained LSO-RF model; The calculation steps of the Spearman rank correlation coefficient are as follows: (1) First, sort the data of the two variables (X, Y) separately, and then record the position after sorting; at this time, the position of a certain data i is X i or Y i rank; (2) Then the Spearman correlation coefficient r s Calculated as follows: where d i For X i With Y i The rank difference between them; n is the number of samples in the sequence; The LSO-RF model is constructed as follows: S41: Population initialization, including initializing the number of lions in the lion group and the position of the lions; S42: Input the RF random forest model and select the parameters to be optimized; S43: Calculate the number of lion kings, lionesses, and lion cubs in the lion group according to the formula, and set the initial group optimal position as the lion king's position, and the current position of each lion as the individual historical optimal position; S44: Set the fitness function of the LSO lion group optimization algorithm to the MAE, RMSE, and R of the RF random forest model prediction results. 2 The average value of S45: Update the lion king's position and calculate the fitness value; S46: Update the position of the lioness and the position of the cubs; S47: Recalculate the fitness value and update the global optimal position and the historical optimal position; S48: Determine whether the number of iterations is the maximum number of iterations. If not, repeat steps S45 to S47. If it is the maximum number of iterations, end the loop and output the value of the parameter to be optimized. S49: Based on the obtained parameters, an LSO-optimized RF random forest model is established.

2. The method for predicting the displacement rate of a step-type landslide based on the LSO-RF model according to claim 1 is characterized in that: Step S1 specifically includes: obtaining the cumulative displacement, pore water pressure, and hourly rainfall of the landslide through a field monitoring station on an hourly basis; processing the cumulative displacement, pore water pressure, and hourly rainfall data and converting them into the landslide displacement rate, daily average pore water pressure, and daily rainfall on a daily basis; and obtaining the cloud cover, humidity, effective soil water holding capacity, and surface runoff in the landslide area through meteorological station data on a daily basis.

3. The method for predicting the displacement rate of a step-type landslide based on the LSO-RF model according to claim 2 is characterized in that: The step S2 is specifically as follows: S21: Consider the landslide displacement rate as the dependent variable Y, and the landslide daily average pore water pressure, daily rainfall, cloud cover, humidity, soil effective water holding capacity, and surface runoff as the independent variables X, and explore the Spearman rank correlation coefficient between X and Y respectively; S22: The landslide displacement rate time series is regarded as the reference time series, and the landslide daily average pore water pressure, daily rainfall, cloud cover, humidity, soil effective water holding capacity, and surface runoff time series are regarded as the comparison time series. After averaging the reference time series and the comparison time series, the grey correlation degree between the reference time series and each comparison time series is calculated. S23: Comprehensively compare the analysis results of Spearman rank correlation coefficient and grey relational degree, and select factors that meet the preset requirements in both analyses as the input features of the model.

4. The method for predicting the displacement rate of a step-type landslide based on the LSO-RF model according to claim 3 is characterized in that: The calculation steps of the grey relational degree are as follows: (1) The time series of landslide displacement rate was determined as the reference series, and the time series of landslide daily average pore water pressure, daily rainfall, cloud cover, humidity, effective soil water holding capacity, and surface runoff were determined as the comparison series; (2) For the reference sequence {x0} and the comparison sequence {x i }, i=1,2,…,5 are dimensionless through averaging. The averaging formula is as follows: Among them, x i (k) is the kth data in a time series; is the average value of a time series; (3) Calculate the grey correlation coefficient between the reference sequence and the comparison sequence. The calculation formula is as follows: Among them, ξ i (k) is the comparison sequence {x i } is the grey correlation coefficient with the reference sequence {x0} at the kth data; ρ is the resolution coefficient; (4) Calculate the grey correlation value between the reference sequence and each comparison sequence according to the following formula:

5. The method for predicting the displacement rate of a step-type landslide based on the LSO-RF model according to claim 1 is characterized in that: The step S4 further uses the conversion of supervised learning data to convert the time series containing different features into a time series with specified input step size and output step size.

6. The method for predicting displacement rate of step-type landslide based on LSO-RF model according to claim 1, characterized in that: When training the model in step S4, the expansion window method is used. (1) First, the training set and the test set are divided into two sets according to a preset ratio. The model is trained on the training set and then predicts the data starting from the starting position in the test set. (2) After the prediction of the above part of the data is completed, the true values ​​of these data are included in the original training set to form a new training set, and then the next part of the data is predicted.

Images