A landslide simulation parameter inversion method based on improved support vector regression

Abstract

The invention discloses a landslide simulation parameter inversion method based on improved support vector regression. The method inverts the data needed in the numerical simulation based on the numerical simulation model and the data collected by on-site monitoring in the target landslide area. The data of a preset landslide simulation parameter type group in the target landslide area is obtained by constructing the landslide simulation parameter inversion model, and the parameters to be optimized of the landslide simulation parameter inversion model are optimized by using the improved support vector regression method. Meanwhile, external conditions constrained by the on-site collected data are added, and the influence of multiple factors in the target landslide area on the data of the preset landslide simulation parameter type group can be comprehensively considered, so that the data of the preset landslide simulation parameter type group is more in line with the actual situation, and the problem that the support vector regression parameters are difficult to determine is solved. The support vector regression prediction error is continuously reduced by optimizing the parameters to be optimized. The method not only has guiding significance for geotechnical parameter inversion, but also has certain research significance in algorithm optimization invention.

Description

Technical Field

[0001] The invention belongs to the field of intelligent numerical simulation, and specifically relates to a landslide simulation parameter inversion method based on improved support vector regression. Background Art

[0002] With the rapid development of computer science and computing, a large number of geotechnical numerical analysis software that can be used to solve practical engineering problems have emerged in the field of geotechnical numerical analysis. These geotechnical numerical analysis software are often used as auxiliary tools for engineering-related decision-making and are used to predict changes related to the engineering construction process. Taking landslide engineering as an example, numerical simulation methods are used to conduct corresponding analysis and research in the assessment of landslide hazards and the design of landslide control plans. It can be seen that the accuracy of numerical simulation results will directly affect the relevant decisions of subsequent projects. Input parameters are an important factor affecting the accuracy of numerical simulation results. At present, numerical simulation parameters often use experimental parameters. In fact, due to the influence of factors such as physical scale and sample disturbance, the parameters obtained from the experiment are quite different from the parameters under natural conditions. With such parameters, the numerical simulation obtained is difficult to provide guidance for engineering.

[0003] Support vector machine is a mature machine learning method, which includes SVM classifier and SVR regression analysis. It is mainly used for classification, recognition and prediction. Among them, SVR regression analysis has been confirmed by a large number of studies to show good adaptability in result prediction and inversion. In SVR regression analysis, there are two parameters that affect the accuracy of prediction results. For different prediction situations, the selection of parameters will be different. How to correctly select SVR parameters has become a problem that needs to be solved, and it is necessary to optimize and improve SVR regression analysis. Summary of the invention

[0004] The technical problem to be solved by the present invention is to solve the problem of selecting input parameters of numerical simulation in the prior art, improve the accuracy of numerical simulation results, and provide accurate judgment for risk assessment and control of landslides.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A landslide simulation parameter inversion method based on improved support vector regression, for a target landslide area, performs the following steps, performs the following steps, constructs a landslide simulation parameter inversion model for the target landslide area, and obtains data of a preset landslide simulation parameter type group for the target landslide area:

[0007] Step A: for the target landslide area, based on the orthogonal numerical simulation method, a landslide numerical model is obtained, which takes the data of the preset landslide simulation parameter type group as input and the displacement data of the target landslide area within a preset time period as output;

[0008] Step B: Based on the monitoring points at each preset position in the target landslide area, the preset time segment is divided into each time period by the preset time step, and based on the landslide numerical model, different data corresponding to the preset landslide simulation parameter type group in each time period and the displacement data of each monitoring point corresponding to the different data are obtained as a training set, and the data of the preset landslide simulation parameter type group in the target landslide area in each time period and the displacement data corresponding to each monitoring point are collected as a test set;

[0009] Step C: For the training set and the test set within the preset time period, the displacement data of each monitoring point in a time period is used as input, and the data of the preset landslide simulation parameter type group corresponding to the displacement data of each monitoring point in the time period is used as output. The landslide simulation parameter inversion model is constructed by the improved support vector regression method to obtain the data of the preset landslide simulation parameter type group in the target landslide area.

[0010] As a preferred technical solution of the present invention, the preset landslide simulation parameter type group includes cohesion c and friction angle

[0011] As a preferred technical solution of the present invention, in the step C, the improved support vector regression method optimizes the parameters to be optimized of the landslide simulation parameter inversion model through the following steps C1 to C3, and constructs the landslide simulation parameter inversion model through the optimized parameters to be optimized:

[0012] Step C1: The parameters to be optimized of the landslide simulation parameter inversion model are set as population individuals, and a prey matrix Prey composed of the parameters to be optimized is randomly generated, where Prey is an n×d matrix, n is the number of populations, and d is the population dimension, i.e., the number of parameters to be optimized; based on the various population individuals in the prey matrix Prey, a landslide simulation parameter inversion model corresponding to each population individual is established, and the error function of the landslide simulation parameter inversion model is used as the fitness function, and the population individual corresponding to the minimum fitness function value is retained, and the population individual is replicated n times to form a predator matrix Elite, and the predator matrix Elite has the same dimension as the prey matrix Prey;

[0013] Step C2: Based on the prey matrix Prey and the predator matrix Elite, for the training set and the test set, combined with the preset number of iterations M, iteratively perform the following process to iteratively update the predator matrix Elite, and finally output the predator matrix Elite:

[0014] Update the prey matrix Prey. Based on the individuals of various populations in the prey matrix Prey, establish an inversion model for landslide simulation parameters corresponding to each individual of various populations. Take the error function of the landslide simulation parameter inversion model as the fitness function. Compare the fitness function value corresponding to the individual of the prey matrix population at the same position with the fitness function value corresponding to the individual of the predator matrix population. If the fitness function value corresponding to the individual of the prey matrix population is better than the fitness function value corresponding to the individual of the predator matrix population, then replace the individual of the predator matrix population with the individual of the prey matrix population, update the predator matrix Elite, and enter the next iteration, repeating step C2; if the fitness function value corresponding to the individual of the prey matrix population is not better than the fitness function value corresponding to the individual of the predator matrix population, then the predator matrix Elite remains unchanged and enters the next iteration, repeating step C2;

[0015] Step C3: For the predator matrix Elite obtained in step C2, screen the individuals of various populations in the predator matrix Elite to obtain an individual of a population, which is the parameter to be optimized obtained by optimization.

[0016] As a preferred technical solution of the present invention, the fitness function F of the landslide simulation parameter inversion model is as follows:

[0017] F=(X′ - X)+(Y′ - Y)

[0018] Wherein, X and Y represent the data of the known preset landslide simulation parameter type group, and X′ and Y′ represent the data of the preset landslide simulation parameter type group obtained through the landslide simulation parameter inversion model.

[0019] As a preferred technical solution of the present invention, at the beginning of each iteration in step C2, the prey matrix Prey is updated by the following method

[0020] When m < M / 3, the prey matrix Prey is updated by the following formula:

[0021]

[0022]

[0023] Wherein, m represents the current iteration number, stepsize i is the moving step size of the i-th individual of the population; R B is a random vector generated by Brownian random walk, and the dimension is d; Prey i is the i-th individual of the population in the prey matrix Prey; Elite i is the i-th individual of the population in the predator matrix Elite; P is a preset constant; R is a vector composed of random numbers uniformly distributed between 0 and 1, and the dimension is d;

[0024] When M / 3≤m<2M / 3, the prey matrix Prey is updated by the following formula:

[0025] The first n / 2 population individuals of the prey matrix Prey are updated by the following formula:

[0026]

[0027]

[0028] The last n / 2 population individuals of the prey matrix Prey are updated by the following formula:

[0029]

[0030]

[0031] Among them, R L is a random vector generated by Levy motion, with dimension d; CF is the step size stepsize i Preset adaptive parameters of

[0032] When 2M / 3≤m, the prey matrix Prey is updated by the following formula:

[0033]

[0034]

[0035] As a preferred technical solution of the present invention, for the prey matrix Prey updated at the beginning of each iteration, the PWLCM chaotic mapping is used to perform chaotic perturbations on the individuals of various populations in the prey matrix Prey, and the fitness function values ​​corresponding to the individuals of the populations before and after the perturbation are compared, and the individuals of the populations with better fitness function values ​​are retained, and the prey matrix Prey is updated for this iteration;

[0036] The mapping formula is:

[0037] E′=E×z m

[0038] The perturbation factor is expressed as:

[0039]

[0040] Among them, E′ is the population individual in the prey matrix Prey after chaotic perturbation, E is the population individual in the prey matrix Prey before mapping, m refers to the number of iterations, z 1 =0.5.

[0041] As a preferred technical solution of the present invention, after each iteration process is completed, the FADs effect is performed on the prey matrix Prey in this iteration process, and the prey matrix Prey is updated to enter the next iteration.

[0042] Update the prey matrix Prey by the following formula:

[0043]

[0044] Among them, FADs is a preset constant, U is a random binary array with dimension d, X min is the lower limit of the search space of the parameters to be optimized, X max is the upper limit of the search space of the parameter to be optimized, r is a random number between [0,1], Prey a Prey b Hunting

[0045] , two random individuals of the population in the object matrix.

[0046] As a preferred technical solution of the present invention, the parameters to be optimized are the regularization coefficient r and the kernel function coefficient σ.

[0047] As a preferred technical solution of the present invention, the process of screening the individuals of various groups in the predator matrix Elite in step C3 is as follows:

[0048] Step C3.1: Based on the predator matrix Elite, according to the fitness function values ​​corresponding to the individuals of each population in the predator matrix Elite, the minimum fitness function value is used as the optimal selection criterion, and a preset number of population individuals are retained;

[0049] Step C3.2: For a preset number of population individuals, use PWLCM chaotic mapping to perform chaotic perturbations on each population individual, compare the fitness function values ​​of the population individuals before and after the chaotic perturbation, retain the population individuals with better fitness function values, and update the preset number of population individuals;

[0050] The mapping formula is:

[0051] E′=E×z M

[0052] The perturbation factor is expressed as:

[0053]

[0054] Among them, E′ is the population individual in the prey matrix Prey after chaotic perturbation, E is the population individual in the prey matrix Prey before mapping, m refers to the number of iterations, z 1 =0.5;

[0055] Step C3.3: Construct the fitting linear equation based on the data of the preset landslide simulation parameter type group collected in the target landslide area For a preset number of population individuals, landslide simulation parameter inversion models corresponding to each population individual are established, and the data of the preset landslide simulation parameter type group in the target landslide area are obtained by using each landslide simulation parameter inversion model. The shortest vertical distance from the point corresponding to the data of each preset landslide simulation parameter type group to the line corresponding to the fitting linear equation is used as the screening principle, and the population individual corresponding to the shortest distance is retained as the optimized parameter to be optimized.

[0056] The beneficial effects of the present invention are as follows: the present invention provides a landslide simulation parameter inversion method based on improved support vector regression, based on a numerical simulation model, combined with data collected by on-site monitoring of a target landslide area to invert the landslide simulation parameters needed for numerical simulation, in the present invention, by constructing a landslide simulation parameter inversion model to obtain data of a preset landslide simulation parameter type group in the target landslide area, the parameters to be optimized of the landslide simulation parameter inversion model are optimized by an improved support vector regression method, and at the same time, external conditions constrained by on-site collected data are added, so that the influence of multiple factors in the target landslide area on the data of the preset landslide simulation parameter type group can be comprehensively considered, so that the data of the preset landslide simulation parameter type group is more in line with the actual situation, and the problem that the support vector regression parameters are difficult to determine is solved, and the support vector regression prediction error is continuously reduced by optimizing the parameters to be optimized, which not only has guiding significance for geotechnical parameter inversion, but also has certain research significance in algorithm optimization invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a flow chart of the method of the present invention;

[0058] Figure 2 A flowchart for optimizing the parameters to be optimized; a block diagram of the improved support vector regression calculation of the present invention;

[0059] Figure 3 It is a numerical model of landslide;

[0060] Figure 4 Output results for landslide numerical models;

[0061] Figure 5 This is the linear equation curve of geotechnical parameters in the field test of the landslide part. DETAILED DESCRIPTION

[0062] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments may enable those skilled in the art to more fully understand the present invention, but are not intended to limit the present invention in any way.

[0063] Since the current landslide simulation parameters are often obtained from experiments, in fact, due to the influence of factors such as physical scale and sample disturbance, the parameters obtained from the experiment are quite different from the parameters under natural conditions. In order to solve the problem of selecting numerical simulation input parameters in the existing technology, improve the accuracy of numerical simulation results, and provide accurate judgment for landslide risk assessment and management, a landslide simulation parameter inversion method based on improved support vector regression is proposed to minimize the difference between simulation results and actual measured values. The improved support vector machine theory is used to optimize the numerical simulation input parameters so that the simulation results are close to the actual measured values.

[0064] A landslide simulation parameter inversion method based on improved support vector regression is performed for the target landslide area, such as Figure 1 As shown, a landslide simulation parameter inversion model of the target landslide area is constructed to obtain data of a preset landslide simulation parameter type group in the target landslide area, wherein the preset landslide simulation parameter type group includes cohesion c and friction angle

[0065] Step A: For the target landslide area, a landslide numerical model is obtained based on the orthogonal numerical simulation method, which takes the data of the preset landslide simulation parameter type group as input and the displacement data of the target landslide area within a preset time period as output; the landslide numerical model should contain model nodes corresponding to the location of the on-site monitoring point, i.e., the monitoring points at each preset location, such as Figure 3 shown.

[0066] Step B: Based on the monitoring points at each preset position in the target landslide area, the preset time segment is divided into each time period by a preset time step, and based on the landslide numerical model, different data corresponding to the preset landslide simulation parameter type group in each time period and the displacement data of each monitoring point corresponding to the different data are obtained as a training set, and the data of the preset landslide simulation parameter type group in the target landslide area in each time period and the displacement data of each monitoring point are collected as a test set; since the range of the landslide parameter interval is too large, in this embodiment, based on the situation of the target landslide area, the data interval of the preset landslide simulation parameter type group obtained by the test in the target landslide area is used as the input parameter interval of the numerical simulation model, and the parameter interval c: [68.10, 80.90] obtained by the test is obtained at a certain interval, Select the input parameters of the landslide model, perform orthogonal numerical simulation, obtain a series of orthogonal numerical simulation results, record each numerical simulation result, and form a learning sample. The numerical simulation result corresponding to the data of one of the preset landslide simulation parameter type groups is as follows: Figure 4As shown, the results include the displacement information of the model nodes corresponding to the monitoring points at each preset location on site; the displacement data of the monitoring points at each preset location corresponding to the data of each preset landslide simulation parameter type group collected on site are used as test samples, namely, training set and test set. In this embodiment, the training set and test set should be no less than 200 groups. Some results are shown in Table 1.

[0067] Table 1. Numerical simulation results of some landslides

[0068]

[0069] Step C: For the training set and the test set within the preset time period, the displacement data of each monitoring point in a time period is used as input, and the data of the preset landslide simulation parameter type group corresponding to the displacement data of each monitoring point in the time period is used as output. The landslide simulation parameter inversion model is constructed by the improved support vector regression method to obtain the data of the preset landslide simulation parameter type group in the target landslide area.

[0070] In step C, the improved support vector regression method is performed through the following steps C1 to C3, such as Figure 2 As shown, the parameters to be optimized of the landslide simulation parameter inversion model are optimized, and the landslide simulation parameter inversion model is constructed by optimizing the parameters to be optimized, wherein the parameters to be optimized are the regularization coefficient r and the kernel function coefficient σ:

[0071] Step C1: Set the parameters to be optimized of the landslide simulation parameter inversion model as population individuals, randomly generate a prey matrix Prey composed of the parameters to be optimized, Prey is an n×d matrix, n is the number of populations, d is the population dimension, i.e., the number of parameters to be optimized, and set the initial parameters of the landslide simulation parameter inversion model, including kernel function, number of iterations, maximum tolerance and other parameters. Determine the search space of the SVR parameters to be optimized (r, σ), and the parameter intervals used in this example are: r: [0.01, 100], σ: [0.01, 100]; Based on the various population individuals in the prey matrix Prey, establish the landslide simulation parameter inversion model corresponding to each population individual, input the displacement data in the data prepared in step B, use the error function of the landslide simulation parameter inversion model as the fitness function, retain the population individual corresponding to the minimum fitness function value, and replicate the population individual n times to form a predator matrix Elite, the predator matrix Elite has the same dimension as the prey matrix Prey.

[0072] Initialize the prey matrix Prey. The Prey matrix is ​​established according to the following expression:

[0073]

[0074] This process mimics the foraging activities of marine organisms. Marine predators choose the optimal foraging strategy between Lévy walks or Brownian walks. In each optimization iteration, the prey selects a walking pattern based on the speed ratio with the predator and updates its own position. The predator updates its optimal foraging position according to the prey's position. The iterative process is repeated until the predator's position meets the condition requirements.

[0075] Step C2: Based on the prey matrix Prey and the predator matrix Elite, for the training set and the test set, combined with the preset number of iterations M, iteratively execute the following process to iteratively update the predator matrix Elite, and finally output the predator matrix Elite: This iterative training process includes a landslide simulation parameter inversion model constructed based on the training set and calibrates it through the test set until the preset number of iterations M is reached, and the iteration ends, outputting the predator matrix Elite:

[0076] Update the prey matrix Prey. Based on the individuals of each population in the prey matrix Prey, establish landslide simulation parameter inversion models corresponding to each population individual respectively. Take the error function of the landslide simulation parameter inversion model as the fitness function, and compare the fitness function value corresponding to the prey matrix population individual at the same position with the fitness function value corresponding to the population individual in the predator matrix. If the fitness function value corresponding to the prey matrix population individual is better than the fitness function value corresponding to the predator matrix population individual, then replace the predator matrix population individual with the prey matrix population individual, update the predator matrix Elite, and enter the next iteration, repeating step C2; if the fitness function value corresponding to the prey matrix population individual is not better than the fitness function value corresponding to the predator matrix population individual, then the predator matrix Elite remains unchanged and enters the next iteration, repeating step C2;

[0077] The fitness function F of the landslide simulation parameter inversion model is as follows:

[0078] F = (X′ - X) + (Y′ - Y)

[0079] Where X and Y represent the data of the known preset landslide simulation parameter type group, that is, the data in the training set and the test set, and X′ and Y′ represent the data of the preset landslide simulation parameter type group obtained through the landslide simulation parameter inversion model.

[0080] At the beginning of each iteration in step C2, update the prey matrix Prey through the following method;

[0081] When m < M / 3, which mainly occurs in the early stage of the optimization iteration process, at this time the prey speed is greater than the predator speed, and the prey position update adopts the Brownian walk method. Update the prey matrix Prey through the following formula:

[0082]

[0083]

[0084] Among them, m represents the current number of iterations, stepsize i is the moving step length of the i-th population individual; R B is a random vector generated by Brownian random walk, with dimension d; Prey i is the i-th individual in the prey matrix Prey; Elite i is the i-th individual in the predator matrix Elite; P is a preset constant; R is a vector of uniformly distributed random numbers between 0 and 1, with a dimension of d;

[0085] When M / 3≤m<2M / 3, it mainly occurs in the middle of the optimization process. At this time, the prey and predator have the same speed, and the population is divided into two parts. The prey performs Lévy motion and is responsible for the algorithm to develop in the search space. The predator performs Brownian motion and is responsible for the algorithm to explore in the search space. The prey matrix Prey is updated by the following formula:

[0086] The first n / 2 population individuals of the prey matrix Prey are updated by the following formula:

[0087]

[0088]

[0089] The last n / 2 population individuals of the prey matrix Prey are updated by the following formula:

[0090]

[0091]

[0092] Among them, R L is a random vector generated by Levy motion, with dimension d; CF is the step size stepsize i Preset adaptive parameters of

[0093] When 2M / 3≤m, it mainly occurs in the late stage of the optimization process. The predator moves faster than the prey. The algorithm mainly focuses on local development and updates the prey matrix Prey through the following formula:

[0094]

[0095]

[0096] For the prey matrix Prey updated at the beginning of each iteration, the PWLCM chaotic mapping is used to perform chaotic perturbations on the individuals of various populations in the prey matrix Prey. In order to weaken the pseudo-randomness of the random number generation process of computer technology itself and further enhance the search capability, the fitness function values ​​corresponding to the individuals of the population before and after the perturbation are compared, and the individuals of the population with better fitness function values ​​are retained, and the prey matrix Prey is updated for this iteration;

[0097] The mapping formula is:

[0098] E′=E×z m

[0099] The perturbation factor is expressed as:

[0100]

[0101] Among them, E′ is the population individual in the prey matrix Prey after chaotic perturbation, E is the population individual in the prey matrix Prey before mapping, m refers to the number of iterations, z 1 =0.5.

[0102] After each iteration, the FADs effect is performed on the prey matrix Prey in this iteration, and the prey matrix Prey is updated to enter the next iteration, creating some opportunities for longer transitions to jump out of the local optimal solution of the algorithm.

[0103] Update the prey matrix Prey by the following formula:

[0104]

[0105] Among them, FADs is a preset constant, U is a random binary array with dimension d, X min is the lower limit of the search space of the parameters to be optimized, X max is the upper limit of the search space of the parameter to be optimized, r is a random number between [0,1], Prey a 、Prey b They are two random population individuals in the prey matrix, where FADs=0.2.

[0106] Step C3: For the predator matrix Elite obtained in step C2, various population individuals in the predator matrix Elite are screened to obtain a population individual, which is the optimized parameter to be optimized.

[0107] The process of screening the individuals of various groups in the predator matrix Elite in step C3 is as follows:

[0108] Step C3.1: Based on the predator matrix Elite, according to the fitness function values ​​corresponding to the individuals in each population in the predator matrix Elite, the minimum fitness function value is used as the optimal selection criterion, and a preset number of population individuals are retained. In this embodiment, the five population individuals with the best fitness are retained.

[0109] Step C3.2: For a preset number of population individuals, use PWLCM chaotic mapping to perform chaotic perturbations on each population individual, compare the fitness function values ​​of the population individuals before and after the chaotic perturbation, retain the population individuals with better fitness function values, and update the preset number of population individuals;

[0110] The mapping formula is:

[0111] E′=E×z M

[0112] The perturbation factor is expressed as:

[0113]

[0114] Among them, E′ is the population individual in the prey matrix Prey after chaotic perturbation, E is the population individual in the prey matrix Prey before mapping, m refers to the number of iterations, z 1 =0.5

[0115] Step C3.3: Consider the correlation between the landslide geotechnical parameters cohesion and internal friction angle in a linear relationship, and construct a fitting linear equation based on the data of the preset landslide simulation parameter type group collected in the target landslide area. For a preset number of population individuals, landslide simulation parameter inversion models corresponding to each population individual are established, and the data of the preset landslide simulation parameter type group in the target landslide area are obtained by using each landslide simulation parameter inversion model. The shortest vertical distance from the point corresponding to the data of each preset landslide simulation parameter type group to the line corresponding to the fitting linear equation is used as the screening principle, and the population individual corresponding to the shortest distance is retained as the optimized parameter to be optimized.

[0116] By using the following distance formula, we can determine the distances of the five population individuals to the linear equation. Distance between:

[0117]

[0118] where c 0 , are the landslide simulation parameters obtained from various population individuals.

[0119] In this embodiment, the landslide simulation parameters collected on site and some landslide field data are used, as shown in Table 2; the fitting linear equation like Figure 5 shown.

[0120] Table 2 Part of the landslide site data

[0121]

[0122] According to the following formula, the five population individuals are determined to the linear equation Distance between:

[0123]

[0124] The input data of the landslide parameter inversion prediction model in this technical solution is obtained from real-time monitoring on site. The monitoring data should take into account timeliness, and the time history of the monitoring data should be the same as the time history of the sample data used to train the inversion model. Based on the construction of the landslide simulation parameter inversion model by the improved support vector regression method, the displacement monitoring data of the monitoring points at each preset location on site are input to obtain the data of the preset landslide simulation parameter type group, that is, the landslide simulation parameters of the target landslide area in the future time direction.

[0125] The invention designs a landslide simulation parameter inversion method based on improved support vector regression. The landslide simulation parameters required for numerical simulation are inverted based on a numerical simulation model and combined with data collected by on-site monitoring of a target landslide area. The theoretical process is solid and feasible. In the invention, a landslide simulation parameter inversion model is constructed to obtain data of a preset landslide simulation parameter type group in a target landslide area. The parameters to be optimized of the landslide simulation parameter inversion model are optimized by an improved support vector regression method. Meanwhile, external conditions constrained by on-site collected data are added. The influence of multiple factors in the target landslide area on the data of the preset landslide simulation parameter type group can be comprehensively considered, so that the data of the preset landslide simulation parameter type group is more in line with the actual situation. The problem that the support vector regression parameters are difficult to determine is solved. The support vector regression prediction error is continuously reduced by optimizing the parameters to be optimized. The invention not only has guiding significance for geotechnical parameter inversion, but also has certain research significance in algorithm optimization invention.

[0126] Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention. A person skilled in the art of the present invention may make various modifications and alterations without departing from the spirit and scope of the present invention.

[0127] The above are only preferred embodiments of the present invention, but do not limit the patent scope of the present invention. Although the present invention is described in detail with reference to the above embodiments, those skilled in the art can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features with equivalent ones. Any equivalent structure made by using the contents of the present invention specification and drawings, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present invention.

Claims

1. A landslide simulation parameter inversion method based on improved support vector regression, characterized in that: For the target landslide area, perform the following steps to construct the landslide simulation parameter inversion model of the target landslide area and obtain the data of the preset landslide simulation parameter type group of the target landslide area: Step A: for the target landslide area, based on the orthogonal numerical simulation method, a landslide numerical model is obtained, which takes the data of the preset landslide simulation parameter type group as input and the displacement data of the target landslide area within a preset time period as output; Step B: Based on the monitoring points at each preset position in the target landslide area, the preset time segment is divided into each time period by the preset time step, and based on the landslide numerical model, different data corresponding to the preset landslide simulation parameter type group in each time period and the displacement data of each monitoring point corresponding to the different data are obtained as a training set, and the data of the preset landslide simulation parameter type group in the target landslide area in each time period and the displacement data corresponding to each monitoring point are collected as a test set; Step C: for the training set and the test set within the preset time period, the displacement data of each monitoring point in a time period is used as input, and the data of the preset landslide simulation parameter type group corresponding to the displacement data of each monitoring point in the time period is used as output, and the landslide simulation parameter inversion model is constructed by the improved support vector regression method to obtain the data of the preset landslide simulation parameter type group in the target landslide area; The preset landslide simulation parameter type group includes cohesion c and friction angle In step C, the improved support vector regression method optimizes the parameters to be optimized of the landslide simulation parameter inversion model through the following steps C1 to C3, and constructs the landslide simulation parameter inversion model through the optimized parameters to be optimized: Step C1: The parameters to be optimized of the landslide simulation parameter inversion model are set as population individuals, and a prey matrix Prey composed of the parameters to be optimized is randomly generated, where Prey is an n×d matrix, n is the number of populations, and d is the population dimension, i.e., the number of parameters to be optimized; based on the various population individuals in the prey matrix Prey, a landslide simulation parameter inversion model corresponding to each population individual is established, and the error function of the landslide simulation parameter inversion model is used as the fitness function, and the population individual corresponding to the minimum fitness function value is retained, and the population individual is replicated n times to form a predator matrix Elite, and the predator matrix Elite has the same dimension as the prey matrix Prey; Step C2: Based on the prey matrix Prey and the predator matrix Elite, for the training set and the test set, combined with the preset number of iterations M, iteratively perform the following process to iteratively update the predator matrix Elite, and finally output the predator matrix Elite: Update the prey matrix Prey, and based on the individuals of various populations in the prey matrix Prey, establish the landslide simulation parameter inversion model corresponding to each individual population, take the error function of the landslide simulation parameter inversion model as the fitness function, and compare the fitness function value corresponding to the individual of the prey matrix population with the minimum fitness function value with the fitness function value corresponding to the individual of the population at the same position in the predator matrix. If the fitness function value corresponding to the individual of the prey matrix population is better than the fitness function value corresponding to the individual of the predator matrix population, then replace the individual of the predator matrix population with the individual of the prey matrix population, update the predator matrix Elite, enter the next iteration, and repeat step C2; if the fitness function value corresponding to the individual of the prey matrix population is not better than the fitness function value corresponding to the individual of the predator matrix population, then the predator matrix Elite remains unchanged and enters the next iteration, and repeats step C2; Step C3: For the predator matrix Elite obtained in Step C2, screen the various population individuals in the predator matrix Elite to obtain a population individual, which is the parameter to be optimized obtained by optimization.

2. The landslide simulation parameter inversion method based on improved support vector regression according to claim 1 is characterized in that: The fitness function F of the landslide simulation parameter inversion model is as follows: F = (X′ - X) + (Y′ - Y) Where X and Y represent the data of the known preset landslide simulation parameter type group, and X′ and Y′ represent the data of the preset landslide simulation parameter type group obtained through the landslide simulation parameter inversion model.

3. The landslide simulation parameter inversion method based on improved support vector regression according to claim 1 is characterized in that: In Step C2, the prey matrix Prey is updated by the following method at the beginning of each iteration. When m < M / 3, the prey matrix Prey is updated by the following formula: Among them, m represents the current number of iterations, stepsize i is the moving step length of the i-th population individual; R B is a random vector generated by Brownian random walk, with dimension d; Prey i is the i-th individual in the prey matrix Prey; Elite i is the i-th individual in the predator matrix Elite; P is a preset constant; R is a vector of uniformly distributed random numbers between 0 and 1, with a dimension of d; When M / 3 ≤ m < 2M / 3, the prey matrix Prey is updated by the following formula: The first n / 2 population individuals of the prey matrix Prey are updated by the following formula: The last n / 2 population individuals of the prey matrix Prey are updated by the following formula: Among them, R L is a random vector generated by Levy motion, with dimension d; CF is the step size stepsize i Preset adaptive parameters of When 2M / 3 ≤ m, the prey matrix Prey is updated by the following formula:

4. The landslide simulation parameter inversion method based on improved support vector regression according to claim 1 is characterized in that: For the prey matrix Prey updated at the beginning of each iteration, chaotic perturbation is performed on the various population individuals in the prey matrix Prey using the PWLCM chaotic mapping. Compare the fitness function values corresponding to the population individuals before and after the perturbation, retain the population individuals with better fitness function values, and update the prey matrix Prey for this iteration; The mapping formula is: E′=E×z m The perturbation factor is expressed as: Where E′ is the population individual after chaotic perturbation of the population individual in the prey matrix Prey, E is the population individual in the prey matrix Prey before mapping, m refers to the iteration number, and z1 = 0.

5.

5. The landslide simulation parameter inversion method based on improved support vector regression according to claim 1 is characterized in that: After each iteration process ends, perform the FADs effect on the prey matrix Prey in this iteration process, update the prey matrix Prey and enter the next iteration. The prey matrix Prey is updated by the following formula: Among them, FADs is a preset constant, U is a random binary array with dimension d, X min is the lower limit of the search space for the parameters to be optimized, X max is the upper limit of the search space of the parameter to be optimized, r is a random number between [0,1], Prey a 、Prey b are two random population individuals in the prey matrix.

6. The landslide simulation parameter inversion method based on improved support vector regression according to claim 1 is characterized in that: The parameter to be optimized is the regularization coefficient r and the kernel function coefficient σ.

7. The landslide simulation parameter inversion method based on improved support vector regression according to claim 1 is characterized by: The process of screening the various population individuals in the predator matrix Elite in Step C3 is as follows: Step C3.1: Based on the predator matrix Elite, according to the fitness function values corresponding to the various population individuals in the predator matrix Elite, with the minimum fitness function value as the optimal selection criterion, retain a preset number of population individuals; Step C3.2: For the preset number of population individuals, perform chaotic perturbation on each population individual using the PWLCM chaotic mapping. Compare the fitness function values corresponding to the population individuals before and after the chaotic perturbation, retain the population individuals with better fitness function values, and update the preset number of population individuals; The mapping formula is: E′=E×z M The perturbation factor is expressed as: Where E′ is the population individual after chaotic perturbation of the population individual in the prey matrix Prey, E is the population individual in the prey matrix Prey before mapping, m refers to the iteration number, and z1 = 0.5; Step C3.3: Construct the fitting linear equation based on the data of the preset landslide simulation parameter type group collected in the target landslide area For a preset number of population individuals, landslide simulation parameter inversion models corresponding to each population individual are established, and the data of the preset landslide simulation parameter type group in the target landslide area are obtained by using each landslide simulation parameter inversion model. The shortest vertical distance from the point corresponding to the data of each preset landslide simulation parameter type group to the line corresponding to the fitting linear equation is used as the screening principle, and the population individual corresponding to the shortest distance is retained as the optimized parameter to be optimized.

Images