Rainfall contour surface generation method based on parameter optimization support vector machine regression

Abstract

The application provides a rainfall isoline generation method and system based on a new parameter optimization support vector machine regression, a medium electronic device, and the like, and includes the following steps: constructing a support vector machine regression model based on a Laplacian kernel, a Gaussian kernel, and a Sigmoid kernel, respectively; performing parameter optimization training on the support vector machine regression models based on the Gaussian kernel, the Laplacian kernel, and the Sigmoid kernel, respectively, to obtain an optimized corresponding parameter combination; updating the support vector machine regression models based on the Gaussian kernel, the Laplacian kernel, and the Sigmoid kernel according to the optimized corresponding parameter combination, respectively, to obtain an optimal model, then inputting point data to be interpolated, performing optimal model interpolation, obtaining an interpolation result of a corresponding point to be interpolated, and then outputting the interpolation result and an isoline map, and saving an optimal parameter and an error of a corresponding interpolation point. The application enhances the representation ability of nonlinear mapping, and improves the model expansibility and robustness.

Description

Technical Field

[0001] This invention relates to the field of rainfall analysis technology, specifically to a method, system, medium, and electronic device for generating rainfall isosurfaces based on parameter-optimized support vector machine regression. Background Technology

[0002] Real-time monitoring and acquisition of water flow plays an important role in flood control and disaster relief. It mainly depends on the changes in real-time rainfall. Each rainfall observation station can summarize the rainfall at the corresponding location in real time. Based on the rainfall spatial interpolation model, the estimated rainfall at non-observation stations can be obtained, and then the areal rainfall in the corresponding area can be obtained. Based on this, the spatial distribution of rainfall can be obtained.

[0003] Common methods for obtaining the spatial distribution of rainfall include the arithmetic mean method, Thiessen polygon interpolation, Kriging interpolation, IDW interpolation, and grid interpolation. The arithmetic mean method cannot accurately reflect rainfall distribution when station distribution is irregular; the Thiessen polygon interpolation does not consider the influence of elevation and is not suitable for areas with large topographic variations; Kriging interpolation is based on the assumption of an inherent stationary process, resulting in limited generalization; the grid interpolation method is based on regional gridding, using measured data to spatially interpolate rainfall values ​​at grid points, thereby obtaining areal rainfall, and the quality of the interpolation result depends on the spatial interpolation. Therefore, the quality of the spatial interpolation method determines the accuracy of rainfall spatial distribution estimation. Models based on assumed prior information have poor adaptability due to the influence of complex and variable weather conditions; although they can simulate some surface changes, their overall performance is poor. Support vector machine (SVM) regression interpolation models use kernel functions for high-dimensional mapping, offering advantages such as small regression errors and strong model representation capabilities. The main factors determining the quality of SVM regression models are the selection of the kernel function and the values ​​of the corresponding parameters. The quality of a support vector machine regression model is determined by the choice of kernel function. A single-kernel support vector machine regression model has limited mapping and representation capabilities and poor robustness.

[0004] Patent document CN109657361A discloses a gas content prediction method, terminal device, and computer-readable storage medium. The method includes: constructing a gas content prediction model based on gas-related parameters; initializing the values ​​of a first predetermined number of input vectors, and determining the optimal input vector based on the values ​​of the input vectors; dividing all other input vectors except the optimal input vector into two groups; calculating the new value corresponding to each input vector in the first group and each input vector in the second group respectively; performing an optimization operation on all new values ​​in the two groups, and updating the optimal input vector based on the optimization result; assigning the median value of the optimal input vector to the gas content prediction model, training the prediction model with a second predetermined number of sample data to obtain an error value, and continuing to search for the optimal input vector until a preset condition is met and then stopping the training.

[0005] However, the parameter optimization operation in patent document CN109657361 is not applicable to parameter optimization of support vector machine regression models.

[0006] A comparative study of spatial interpolation of precipitation based on BP neural network and support vector machine—taking Gansu Province as an example—was published in Grassland and Turf, 2018, 38(4):8. Using BP neural network and support vector machine models as the research object, Gansu Province was selected as the research area. Spatial interpolation models for precipitation were constructed, and spatial interpolation results for Gansu Province based on the two models were completed. The province was divided into three modeling zones: western, central, and eastern. The accuracy and application differences of the two spatial interpolation models were compared and analyzed. The results showed that the interpolation accuracy of the support vector machine model was significantly higher than that of the BP neural network model, and the support vector machine model could reflect more details in the spatial distribution of precipitation.

[0007] Although the literature also uses support vector machine regression based on latitude, longitude and elevation data for spatial interpolation of rainfall, this invention uses a novel optimization algorithm to optimize the multi-core combined support vector machine regression model, thereby comprehensively improving the optimization accuracy and algorithm robustness.

[0008] In summary, the existing interpolation methods have disadvantages, and the support vector machine regression model has high parameter optimization complexity and is prone to getting trapped in local optima. Summary of the Invention

[0009] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method, system, medium, and electronic device for generating rainfall isosurfaces based on parameter-optimized support vector machine regression.

[0010] A method for generating rainfall isosurfaces based on parameter-optimized support vector machine regression, provided by the present invention, includes:

[0011] Model building steps: Build support vector machine regression models based on the Laplace kernel, Gaussian kernel, and Sigmoid kernel respectively;

[0012] Model parameter optimization steps: Perform parameter optimization training on support vector machine regression models based on Gaussian kernel, Laplace kernel and Sigmoid kernel respectively to obtain the corresponding optimal parameter combinations;

[0013] The steps for generating rainfall isosurfaces are as follows: Based on the optimized parameter combinations, the support vector machine regression models based on Gaussian kernel, Laplace kernel, and Sigmoid kernel are updated respectively to obtain the optimal model. Then, the data of the points to be interpolated are input, the optimal model is used for interpolation, the interpolation results of the corresponding points to be interpolated are obtained, and then the interpolation results and isosurface maps are output. At the same time, the optimal parameters and the errors of the corresponding interpolation points are saved.

[0014] Preferably, the model building step includes:

[0015] Step S1.1: Obtain historical rainfall data as a training sample set, as shown below:

[0016] {(x 1,1 ,y 1,1 ,z 1,1 ,h 1,1 ),...,(x i,1 ,y i,1 ,z i,1 ,h i,1 ),...,(x n,1 ,y n,1 ,z n,1 ,h n,1 ),...,(x 1,t ,y 1,t ,z 1,t ,h 1,t ),..,(x i,t ,y i,t ,z i,t ,h i,t ),...,(x n,t ,y n,t ,z n,t ,h n,t ),...,(x 1,T ,y 1,T ,z 1,T ,h 1,T ),...,(x i,T ,y i,T ,z i,T ,h i,T i = 1, ..., n, ..., (x n,T ,y n,T ,z n,T ,h n,T )}

[0017] t=1,...,T

[0018] Where T represents the total number of statistical periods, n represents the total number of observation stations, and x i,t ,y i,t ,z i,t ,h i,t Let represent the longitude, latitude, elevation, and measured rainfall data within the statistical time period t of the i-th observation station, respectively;

[0019] Step S1.2: Based on the training sample set, determine the nonlinear regression equation for the support vector machine as follows:

[0020]

[0021] In the formula, X represents LInterpolation prediction data for observation points, where φ represents the kernel function, X L Let w represent the interpolation point, and b represent the weight vector and bias, respectively.

[0022] The loss function is:

[0023]

[0024] In the formula, ξ, They are slack variables, and C represents the penalty coefficient, ε represents the deviation, and h i This represents the measured data at each observation point;

[0025] Step S1.3: Solve the nonlinear regression equation of the support vector machine to obtain:

[0026]

[0027] In the formula, α i and Represents the Lagrange multipliers.

[0028] Step S1.4: Based on the solved equations, construct the support vector machine regression model using the Laplace kernel function as follows:

[0029] The support vector machine regression model constructed based on the Gaussian kernel function is as follows:

[0030] The support vector machine regression model constructed based on the Sigmoid kernel function is: k3(x i ,x j )=tanh(βx T i x j +θ)

[0031] In the formula, x i and x j Let represent the i-th and j-th interpolation points respectively, δ represent the Laplace kernel parameter and Gaussian kernel parameter, and β and θ represent the Sigmoid kernel parameter.

[0032] Preferably, the model parameter optimization step includes the following parameter optimization algorithm, which optimizes the regression model parameters of Gaussian kernel support vector machine, Laplace kernel support vector machine and Sigmoid kernel support vector machine respectively:

[0033] Step S2.1: Initialize the corresponding support vector machine regression model parameters and the corresponding novel parameter optimization algorithm parameters; the novel parameter optimization algorithm parameters include the number of individuals in the population cluster corresponding to each model and the maximum number of iterations of the population; the individuals represent the parameter optimization space;

[0034] Step S2.2: Acquire data and divide the data into regions;

[0035] Step S2.3: Perform interpolation training for each region, calculate the initial value of the optimal fitness, and select the best individual within the population cluster;

[0036] Step S2.4: Perform chaotic optimization on all individuals other than the optimal individual, calculate the current optimal fitness value, and select the optimal individual in the current population cluster from the new parameter optimization parameters;

[0037] Step S2.5: Calculate the number of offspring in the current best individual, and calculate the fitness value of the offspring generated by each individual based on the number of offspring. Determine whether the fitness value of the offspring generated by each individual is better than the current best fitness value. If so, update the current best fitness value and the current best individual, and delete the non-optimal individuals generated by each individual to ensure that the total population size remains unchanged; if not, do not update.

[0038] Step S2.6: Divide all individuals in the current population cluster except for the best individual into two groups. Perform a tendency operation on the first group of individuals and an exploration operation on the second group of individuals. Calculate and update the current best fitness value and the best individual.

[0039] Step S2.7: Eliminate the K individuals with the lowest objective function values, and regenerate the K individuals in the solution space according to step S2.1. Then calculate and update the current optimal fitness value and the optimal individual.

[0040] Step S2.8: Determine whether the termination condition has been met. If yes, terminate the output model and trigger step S2.9; otherwise, trigger step S2.2.

[0041] Step S2.9: Output the Gaussian kernel support vector machine regression model, Laplace kernel support vector machine regression model, and Sigmoid kernel support vector machine regression model corresponding to the optimal parameters.

[0042] Preferably, the chaotic optimization in step S2.4 is updated using the following formula:

[0043]

[0044]

[0045] In the formula, Let j represent the j-th individual in the k-th generation of the i-th support vector machine regression model. Let j represent the j-th individual in the (k-1)-th generation of the i-th support vector machine regression model. and Let s represent the individuals with the best and worst fitness values ​​in the (k-1)th generation of each population cluster, respectively. k =μ(1-s k-1 ) represents the k-th generation chaotic operator, where 0 <s<1,0<μ≤4,N max This represents the maximum number of iterations for the population.

[0046] Preferably, the formula for calculating the number of offspring in the current optimal individual in step S2.5 is as follows:

[0047]

[0048]

[0049] In the formula, N i,j Let represent the number of offspring of the j-th individual in the i-th support vector machine regression model. and Let these represent the minimum and maximum fitness values ​​in the current k-th generation within the i-th population cluster, respectively. This represents the position fitness value of the j-th individual in the k-th generation of the i-th support vector machine regression model.

[0050] Preferably, the approach operation includes moving closer to the global optimum using the following formula:

[0051]

[0052] in, Let j represent the j-th individual in the k-th generation of the i-th support vector machine regression model. This represents the individual in the k-th generation within the cluster that has the optimal fitness value. ε represents a smaller value to avoid invalidation of the denominator. Let be the position fitness value of the j-th individual in the k-th generation of the i-th support vector machine regression model. This represents the optimal fitness value of the i-th support vector machine regression model in the k-th generation of the current population cluster;

[0053] The exploration operation includes performing a random walk using the following formula:

[0054]

[0055] in, Let represent the j-th individual in the k-th generation of the i-th support vector machine regression model, r1 and r2 represent random numbers between 0 and 1, G(1.5, r2) represents Gaussian random numbers, and levy represents the Levy flight update value. It represents the orientation factor of the j-th individual in the k-th generation of the i-th support vector machine regression model.

[0056] Preferably, the rainfall isosurface generation step further includes an incremental sample simulation interpolation and model optimization step and an incremental update optimization training step:

[0057] Incremental sample simulation interpolation and model optimization steps: When new sample data is added, the input sample dataset is used for simulation interpolation, and the results of Gaussian kernel support vector machine regression model, Laplace kernel support vector machine regression model, and Sigmoid kernel support vector machine regression model are output in parallel. The model with the smallest fitness value is selected as the optimal model. Then, it is determined whether the optimal fitness value meets the threshold requirement. If so, the model is selected, and the incremental sample interpolation output is updated directly over time, and the incremental sample interpolation results are saved. If not, the incremental update optimization training step is executed.

[0058] Incremental update optimization training steps: Input the current sample into the Gaussian kernel support vector machine interpolation model, the Laplace kernel support vector machine interpolation model, and the Sigmoid kernel support vector machine interpolation model in parallel to optimize the model parameters.

[0059] A rainfall isosurface generation system based on parameter-optimized support vector machine regression, provided by the present invention, includes:

[0060] Model building module: Build support vector machine regression models based on the Laplace kernel, Gaussian kernel, and Sigmoid kernel respectively;

[0061] Model parameter optimization module: Perform parameter optimization training on support vector machine regression models based on Gaussian kernel, Laplace kernel and Sigmoid kernel respectively to obtain the corresponding optimal parameter combination;

[0062] Rainfall isosurface generation module: Based on the optimized parameter combination, update the support vector machine regression model based on Gaussian kernel, Laplace kernel and Sigmoid kernel respectively to obtain the optimal model. Then, input the data of the points to be interpolated, perform optimal model interpolation, obtain the interpolation results of the corresponding points to be interpolated, and then output the interpolation results and isosurface map, while saving the optimal parameters and the error of the corresponding interpolation points.

[0063] According to the present invention, a computer-readable storage medium storing a computer program is provided, wherein when the computer program is executed by a processor, the steps of the method for generating rainfall isosurfaces based on parameter-optimized support vector machine regression are implemented.

[0064] According to the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the computer program is executed by the processor, it implements the steps of the rainfall isosurface generation method based on parameter-optimized support vector machine regression.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] 1. This invention optimizes the support vector machine regression rainfall interpolation model based on a novel parameter optimization method, and supports the construction of a rainfall interpolation model with high accuracy.

[0067] 2. This invention is based on threshold control. Data is only retrained when the fitness value is greater than the threshold, which supports continuous optimization of the model with the continuous increase of rainfall observation data. It supports controllable model updates, reduces the frequency of interpolation results and isosurface map updates, reduces the amount of computation such as rendering, and supports dynamic updates of interpolation results and isosurface maps.

[0068] 3. This invention combines three nonlinear support vector machine regression interpolation models for parallel training, optimization, and interpolation output; it supports comparative interpolation training and optimization using Gaussian kernel, Laplace kernel, and Sigmoid kernel, selects the optimal model, enhances the representation ability of nonlinear mapping, and improves model scalability and robustness. Attached Figure Description

[0069] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0070] Figure 1 This is a schematic diagram of the novel parameter optimization algorithm in this invention.

[0071] Figure 2 This is a schematic diagram of the process for generating spatial isosurfaces of rainfall based on novel parameter-optimized support vector machine regression according to the present invention.

[0072] Figure 3 This is a graph showing the effect of regression interpolation results of the novel parameter-optimized support vector machine in this invention.

[0073] Figure 4 This is an image showing the effect of isosurface generation in this invention. Detailed Implementation

[0074] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0075] This invention utilizes a novel parameter-optimized support vector machine (SVM) regression interpolation model that employs kernel functions for high-dimensional mapping, offering advantages such as low regression error and strong model representation capabilities. By optimizing SVM regression model parameters using this novel parameter-optimized approach, the complexity of the optimization process is reduced, avoiding getting trapped in local optima and enabling the construction of highly accurate rainfall interpolation models. Furthermore, the quality of the kernel function selection directly impacts the performance of the SVM regression model; a single-kernel SVM regression model suffers from limited mapping representation capabilities and poor robustness. Commonly used kernel functions for nonlinear mapping include Gaussian, Laplace, and Sigmoid kernels. This invention supports comparative interpolation training optimization using Gaussian, Laplace, and Sigmoid kernels, selecting the optimal model to enhance the representation capabilities of nonlinear mappings and improve model scalability and robustness.

[0076] The present invention provides a method for generating rainfall isosurfaces based on support vector machine regression with parameter optimization, comprising: a model building step, a model parameter optimization step, and a rainfall isosurface generation step.

[0077] Model construction steps: Support vector machine regression models are constructed based on the Laplace kernel, Gaussian kernel, and Sigmoid kernel, respectively. The model construction steps include:

[0078] Step S1.1: Obtain historical rainfall data as a training sample set, as shown below:

[0079] {(x 1,1 ,y 1,1 ,z 1,1 ,h 1,1 ),...,(x i,1 ,y i,1 ,z i,1 ,h i,1 ),...,(x n,1 ,y n,1 ,z n,1 ,h n,1 ),...,(x 1,t ,y 1,t ,z 1,t ,h 1,t )

[0080] ,..,(x i,t ,y i,t ,z i,t ,hi,t ),...,(x n,t ,y n,t ,z n,t ,h n,t ),...,(x 1,T ,y 1,T ,z 1,T ,h 1,T ),...,(x i,T ,y i,T ,z i,T ,h i,T i = 1, ..., n

[0081] ,...,(x n,T ,y n,T ,z n,T ,h n,T )}

[0082] t=1,...,T

[0083] Where T represents the total number of statistical periods, n represents the total number of observation stations, and x i,t ,y i,t ,z i,t ,h i,t Let represent the longitude, latitude, elevation, and measured rainfall data within the statistical time period t of the i-th observation station, respectively.

[0084] Step S1.2: Based on the training sample set, determine the nonlinear regression equation for the support vector machine as follows:

[0085]

[0086] In the formula, X represents L Interpolation prediction data for observation points, where φ represents the kernel function, X L Let w represent the interpolation point, and b represent the weight vector and bias, respectively.

[0087] The loss function is:

[0088]

[0089] In the formula, ξ, They are slack variables, and C represents the penalty coefficient, ε represents the deviation, and h i This represents the measured data at each observation point.

[0090] Step S1.3: Solve the nonlinear regression equation of the support vector machine to obtain:

[0091]

[0092] In the formula, hi For each observation point, X = (X1,...,X...). nT ), For X L Interpolation prediction data for observation points, X = (X1,...,X2) nT ), α i and Represents the Lagrange multipliers.

[0093] Step S1.4: The kernel function maps a linearly inseparable space to a nonlinear space. The choice of kernel function is crucial for constructing the support vector machine model. Therefore, based on the solved equations, the support vector machine regression model constructed using the Laplace kernel function is as follows:

[0094] The support vector machine regression model constructed based on the Gaussian kernel function is as follows:

[0095] The support vector machine regression model constructed based on the Sigmoid kernel function is: k3(x i ,x j )=tanh(βx T i x j +θ)

[0096] β>0, θ<0

[0097] In the formula, x i and x j Let represent the i-th and j-th values ​​to be interpolated, respectively; δ represents the Gaussian kernel parameter and the Laplace kernel parameter; and β and θ represent the Sigmoid kernel parameter.

[0098] This invention selects the optimal model by training the above models separately and then selecting the optimal model based on the optimal loss function, i.e., min(F(X)). The formula for calculating the optimal fitness, i.e., the overall loss function, is as follows:

[0099]

[0100] In the formula, h i This represents the measured data at the i-th observation point. Interpolation prediction data for the i-th observation point

[0101] Model parameter optimization steps: Perform parameter optimization training on support vector machine regression models based on Gaussian kernel, Laplace kernel, and Sigmoid kernel respectively to obtain the corresponding optimal parameter combinations. Specifically, based on the above three models, it can be seen that for support vector regression models with Gaussian kernel and Laplace kernel, ε, C, and δ need to be determined; for the Sigmoid kernel, ε, C, β, and θ need to be solved.

[0102] Furthermore, such as Figure 1 As shown, the specific steps for optimizing model parameters are described below:

[0103] The model parameter optimization steps include the following parameter optimization algorithms, which optimize the regression model parameters of Gaussian kernel support vector machine, Laplace kernel support vector machine, and Sigmoid kernel support vector machine respectively:

[0104] Step S2.1: Initialize the corresponding support vector machine regression model parameters and the corresponding novel parameter optimization algorithm parameters; the novel parameter optimization algorithm parameters include the number of individuals in the population cluster corresponding to each model and the maximum number of iterations of the population. The individuals represent the parameter optimization space.

[0105] Specifically, a) initialize the parameters W0 of the Gaussian kernel support vector machine, W1 of the Laplace kernel support vector machine, and W2 of the Sigmoid kernel support vector machine, where W0 = (ε0, C0, δ0), W1 = (ε1, C1, δ1), and W2 = (ε2, C2, β, θ). b) initialize the number of N0 individuals in the population cluster corresponding to the new parameter optimization of the Gaussian kernel support vector machine parameters, representing the N0 parameters with different positions. The number of N1 individuals in the population cluster corresponding to the novel parameter optimization for initializing the Laplace kernel support vector machine parameters represents N1 parameters with different positions. The number of N2 individuals in the population cluster corresponding to the novel parameter optimization for initializing the Sigmoid kernel support vector machine parameters represents N2 parameters with different positions. c) Initialize the maximum number of iterations N for the population max Algorithm parameters. Randomly initialize parameters:

[0106] W0=λ(W 0,up -W 0,down )+W 0,down

[0107] W1=λ(W 1,up -W 1,down )+W 1,down

[0108] W2=λ(W 2,up -W2,down )+W 2,down

[0109] First, determine the parameter optimization space, W. i,up Let i = 0, 1, 2 be the upper limit vector of parameters, and W i,down Let i = 0, 1, 2 be the lower bound vector of parameters. Each individual represents the parameter optimization vector space, λ = rand, where rand is a random number between 0 and 1. The fitness value of individuals within the cluster is calculated as follows: The optimal individual is determined based on the optimal fitness value, that is, the individual with the best fitness value.

[0110] Step S2.2: Acquire data and divide the data into regions. Step S2.3: Perform interpolation training on each region, calculate the initial optimal fitness value, and select the best individual within the population cluster.

[0111] Specifically, the terrain is divided into regions, and areas with significantly different terrain features are separated. Interpolation training is performed for each segmented region. To avoid the large terrain differences affecting the interpolation model, historical rainfall data for the corresponding region is obtained. Based on the known dataset of observation points in each time period of the historical rainfall records: X = (X1,...,X...) nT ), {(x 1,1 ,y 1,1 ,z 1,1 ,h 1,1 ),...,(x i,1 ,y i,1 ,z i,1 ,h i,1 ),...,(x n,1 ,y n,1 ,z n,1 ,h n,1 ),...,(x 1,t ,y 1,t ,z 1,t ,h 1,t ),..,(x i,t ,y i,t ,z i,t ,h i,t ),...,(x n,t ,y n,t ,z n,t ,h n,t ),...,(x 1,T ,y 1,T ,z 1,T ,h 1,T ),...,(x i,T ,y i,T ,z i,T ,h i,Ti = 1, ..., n, ..., (x n,T ,y n,T ,z n,T ,h n,T )}t=1,...,T, where T is the total statistical period, n is the total number of observation stations, and x i,t ,y i,t ,z i,t ,h i,t Given the longitude, latitude, elevation, and measured statistical rainfall data within the statistical time period t of the i-th observation station, output the point X to be interpolated. L The forecast results For interpolating and predicting data at each observation point, X L =(x L ,y L ,z L () represents the point to be interpolated. After initialization, the novel parameter optimization algorithm is based on... Calculate the initial optimal fitness value, k = 0. The optimal parameter combination is found based on the following steps.

[0112] Step S2.4: In order to effectively explore the internal space of the cluster, chaotic optimization is performed on all individuals other than the optimal individual, the current optimal fitness value is calculated, and the optimal individual in the current population cluster is selected from the new parameter optimization parameters.

[0113] Specifically, the chaotic optimization in step S2.4 is updated using the following formula:

[0114]

[0115]

[0116] In the formula, Let j represent the j-th individual in the k-th generation of the i-th support vector machine regression model. Let j represent the j-th individual in the (k-1)-th generation of the i-th support vector machine regression model. When i = 0, the Gaussian kernel support vector machine is: j = 1, ..., N0; when i = 1, the Laplace kernel support vector machine is: j = 1, ..., N1; when i = 2, the Sigmoid kernel support vector machine is: j = 1, ..., N2. and Let s represent the individuals with the best and worst fitness values ​​in the (k-1)th generation of each population cluster, respectively. k =μ(1-s k-1 ) represents the k-th generation chaotic operator, where 0 <s<1,0<μ≤4,N max This represents the maximum number of iterations for the population. Generally, μ = 4 is used. Based on... Calculate fitness, update It represents the optimal fitness value of the i-th support vector machine regression model in the k-th generation within the cluster. i = 0, 1, 2.

[0117] Step S2.5: The number of offspring for each individual depends on its current fitness and the best and worst fitness values ​​in the current population. The stronger the individual's fitness, the more offspring it produces in each iteration. Calculate the number of offspring from the current best individual, and calculate the fitness value of each individual's offspring based on the number of offspring. Determine if the fitness value of the offspring is better than the current best fitness value. If so, update the current best fitness value and the current best individual; that is, update... Let these be the minimum and maximum fitness values ​​in the i-th cluster at the k-th generation. Let be the position fitness value of the j-th individual in the k-th generation of the i-th support vector machine regression model, and update . Furthermore, delete the suboptimal individuals generated by each individual's reproduction to ensure that the total population size remains unchanged; otherwise, do not update.

[0118] Specifically, the formula for calculating the number of offspring in the current optimal individual in step S2.5 is as follows:

[0119]

[0120]

[0121] In the formula, N i,j Let represent the number of offspring of the j-th individual in the i-th support vector machine regression model. and Let these represent the minimum and maximum fitness values ​​in the current k-th generation within the i-th population cluster, respectively. This represents the position fitness value of the j-th individual in the k-th generation of the i-th support vector machine regression model. i = 0, 1, 2. When i = 0, it is a Gaussian kernel support vector machine regression model, j = 1, ..., N0. When i = 1, it is a Laplace kernel support vector machine regression model, j = 1, ..., N1. When i = 2, it is a Sigmoid kernel support vector machine regression model, j = 1, ..., N2.

[0122] Step S2.6: Divide all individuals in the current population cluster except for the optimal individual into two groups. Perform a tendency operation on the first group and an exploration operation on the second group. Calculate and update the current optimal fitness value and the optimal individual. In other words, a certain proportion of individuals within the cluster, excluding the optimal individual, undergo adaptive tendency operation, while the remaining individuals fly to explore, simulating their response to the environment. Iterate through the individuals in the cluster sequentially, generating a random number r = rand, r ∈ (0,1). If r > 0.5, the individual is considered an explorer; otherwise, they are considered a tendency operator.

[0123] Specifically, the approach operation involves moving closer to the global optimum using the following formula:

[0124]

[0125] in, Let represent the j-th individual in the k-th generation of the i-th support vector machine regression model. When i = 0, the Gaussian kernel support vector machine is: j = 1,...,N0; when i = 1, the Laplace kernel support vector machine is: j = 1,...,N1; when i = 2, the Sigmoid kernel support vector machine is: j = 1,...,N2. This represents the individual in the k-th generation within the cluster that has the optimal fitness value. ε represents a smaller value to avoid invalidation of the denominator. Let be the position fitness value of the j-th individual in the k-th generation of the i-th support vector machine regression model. Let represent the optimal fitness value of the i-th support vector machine regression model in the k-th generation of the current population. Here, ε is a small value to avoid denominator invalidity; when... Individuals are almost unaffected by poorer individuals; when Individuals will strive to become better individuals. Through the tendency operation, the best individuals in the cluster are continuously driven toward the global optimum, ensuring the algorithm's global optimization.

[0126] The exploration operation includes performing a random walk using the following formula:

[0127]

[0128] in, Let represent the j-th individual in the k-th generation of the i-th support vector machine regression model. When i = 0, the Gaussian kernel support vector machine j is: j = 1,...,N0; when i = 1, the Laplace kernel support vector machine j is: j = 1,...,N1; when i = 2, the Sigmoid kernel support vector machine j is: j = 1,...,N2. r1, r2 represent random numbers between 0 and 1, G(1.5,r2) represents Gaussian random numbers, and levy represents the Levy flight update value. Let represent the orientation factor of the j-th individual in the k-th generation of the i-th support vector machine regression model. The levy is generated as follows:

[0129]

[0130] in:

[0131] 0<τ≤2

[0132]

[0133] σ v =1

[0134] τ is typically taken as 1.5, and u and v are normally distributed random numbers. levy is the Lévy flight update value. levy*G(1.5,r2) combines Lévy flight and Gaussian random process to ensure that the algorithm has a good optimization range.

[0135] The direction factor is determined as follows, for support vector regression models with Gaussian and Laplace kernels:

[0136]

[0137] in, The variable search angle is for the j-th individual in the k-th generation of the i-th support vector machine regression model. When using a Sigmoid kernel:

[0138]

[0139] in, The variable search angle is for the j-th individual in the k-th generation of the i-th support vector machine regression model.

[0140] By performing orientation operations on the first group of individuals and exploration operations on the second group of individuals, the current optimal fitness value and the optimal individual are calculated and updated. Let these be the minimum and maximum fitness values ​​in the i-th cluster at the k-th generation. The position fitness value of the j-th individual in the k-th generation of the i-th support vector machine regression model, and the updated

[0141] Step S2.7: Eliminate the K individuals with the lowest objective function values, and regenerate K individuals in the solution space according to step S2.1. Then calculate and update the current optimal fitness value and the optimal individual. Let these be the minimum and maximum fitness values ​​in the i-th cluster at the k-th generation. Let be the position fitness value of the j-th individual in the k-th generation of the i-th support vector machine regression model, and update .

[0142] Step S2.8: Determine whether the termination condition has been met. If yes, terminate the output model and trigger step S2.9; otherwise, trigger step S2.4.

[0143] Step S2.9: Output the Gaussian kernel support vector machine regression model, Laplace kernel support vector machine regression model, and Sigmoid kernel support vector machine regression model corresponding to the optimal parameters. That is, output the optimal parameters W respectively. 0,best This corresponds to the Gaussian kernel support vector machine regression model; it outputs the optimal parameter W. 1,best This corresponds to the Laplace kernel support vector machine regression model; it outputs the optimal parameter W. 2,best This corresponds to the Sigmoid kernel support vector machine regression model.

[0144] The steps for generating rainfall isosurfaces are as follows: Based on the optimized parameter combinations, the support vector machine regression models based on Gaussian kernel, Laplace kernel, and Sigmoid kernel are updated respectively to obtain the optimal model. Then, the data of the points to be interpolated are input, the optimal model is used for interpolation, the interpolation results of the corresponding points to be interpolated are obtained, and then the interpolation results and isosurface maps are output. At the same time, the optimal parameters and the errors of the corresponding interpolation points are saved.

[0145] Specifically, such as Figure 2 As shown, after the optimization process concludes and the optimal parameter combination is obtained, the parameters of the Gaussian kernel support vector machine (SVM) model, the Laplace kernel SVM model, and the Sigmoid kernel SVM model are updated. Then, the loss function for interpolation of these models is calculated, and the interpolation model corresponding to the model with the smallest fitness value is selected as the optimal model, which is then output. Finally, the data for the points to be interpolated is input, and the optimal model is used for interpolation. The interpolation results for the corresponding points are obtained, and the optimal parameters and the errors of the corresponding interpolation points are saved for subsequent error analysis. The interpolation results and isosurface plots are also output.

[0146] Furthermore, when training data is updated, ordinary support vector machine regression models need to determine whether to perform incremental learning or directly perform full training and update. This invention, based on threshold control, only retrains the data when the fitness value is greater than the threshold. This supports continuous model optimization with continuously increasing rainfall observation data, supports dynamic updates of interpolation results and isosurfaces, supports controllable model updates, reduces the frequency of updates to interpolation results and isosurface maps, and reduces computational load such as rendering. Specifically, the rainfall isosurface generation step also includes incremental sample simulation interpolation and model optimization steps, and incremental update optimization training steps. Finally, based on region truncation and interpolation results, the interpolation results are output, and graphic rendering is performed to generate isosurface maps and rainfall level displays. The corresponding results are then updated.

[0147] Incremental sample simulation interpolation and model optimization steps: When new sample data is added, the input sample dataset is used for simulation interpolation, and the results of Gaussian kernel support vector machine regression model, Laplace kernel support vector machine regression model, and Sigmoid kernel support vector machine regression model are output in parallel. The model with the smallest fitness value is selected as the optimal model. Then, it is determined whether the optimal fitness value meets the threshold requirement. If so, the model is selected, and the incremental sample interpolation output is updated directly over time, and the incremental sample interpolation results are saved. If not, the incremental update optimization training step is executed.

[0148] Incremental update optimization training steps: Input the current sample into the Gaussian kernel support vector machine interpolation model, the Laplace kernel support vector machine interpolation model, and the Sigmoid kernel support vector machine interpolation model in parallel to optimize the model parameters.

[0149] Furthermore, taking the Lengshui River Basin as an example, the method for generating rainfall isosurfaces based on parameter optimization and support vector machine regression according to the present invention is described in detail below:

[0150] First, the interpolation area is divided, and vector map and elevation data of the Lengshui River basin are obtained, as well as vector map and elevation data of the observation stations.

[0151] Then, the parameters of the Gaussian kernel support vector machine regression model, the Laplace kernel support vector machine regression model, and the Sigmoid kernel support vector machine regression model are optimized based on a novel parameter optimization algorithm, and the corresponding optimal parameters are output.

[0152] Next, model optimization is performed. Based on the loss function of interpolation of Gaussian kernel support vector machine model, Laplace kernel support vector machine model and Sigmoid kernel support vector machine model, the interpolation model corresponding to the model with the smallest fitness value is selected as the optimal model, and the best interpolation result and the corresponding optimal model are output.

[0153] Next, input the data for the points to be interpolated, simulate interpolation, obtain the interpolation results for the corresponding points, and save the optimal parameters and the errors of the corresponding interpolation points for subsequent error analysis. The interpolation results are as follows: Figure 3 As shown.

[0154] Then, based on the interpolation results, an isosurface map is generated; based on the precipitation level determination, a precipitation level map is generated. The final result is as follows. Figure 4 As shown.

[0155] Finally, determine if there is any new data, optimize the model, and decide whether to retrain. If necessary, retrain and update the interpolation results as follows. Figure 4 As shown.

[0156] This invention also provides a rainfall isosurface generation system based on parameter-optimized support vector machine regression. Those skilled in the art can implement the rainfall isosurface generation system based on parameter-optimized support vector machine regression by executing the steps of the rainfall isosurface generation method based on parameter-optimized support vector machine regression. That is, the rainfall isosurface generation method based on parameter-optimized support vector machine regression can be understood as a preferred embodiment of the rainfall isosurface generation system based on parameter-optimized support vector machine regression.

[0157] A rainfall isosurface generation system based on parameter-optimized support vector machine regression, provided by the present invention, includes:

[0158] Model building module: Constructs support vector machine regression models based on the Laplace kernel, Gaussian kernel, and Sigmoid kernel, respectively. The model building module includes:

[0159] Module M1.1: Obtain historical rainfall data as a training sample set, as shown below:

[0160] {(x 1,1 ,y 1,1 ,z 1,1 ,h 1,1 ),...,(x i,1 ,y i,1 ,z i,1 ,h i,1 ),...,(x n,1 ,y n,1 ,z n,1 ,h n,1 ),...,(x 1,t ,y 1,t ,z 1,t ,h 1,t ),..,(x i,t ,y i,t ,z i,t ,h i,t),...,(x n,t ,y n,t ,z n,t ,h n,t ),...,(x 1,T ,y 1,T ,z 1,T ,h 1,T ),...,(x i,T ,y i,T ,z i,T ,h i,T i = 1, ..., n, ..., (x n,T ,y n,T ,z n,T ,h n,T )}t=1,...,T

[0161] Where T represents the total number of statistical periods, n represents the total number of observation stations, and x i,t ,y i,t ,z i,t ,h i,t Let represent the longitude, latitude, elevation, and measured rainfall data within the statistical time period t of the i-th observation station, respectively;

[0162] Module M1.2: Based on the training sample set, the nonlinear regression equation for the support vector machine is determined as follows:

[0163]

[0164] In the formula, X represents L Interpolation prediction data for observation points, where φ represents the kernel function, X L Let w represent the interpolation point, and b represent the weight vector and bias, respectively.

[0165] The loss function is:

[0166]

[0167] In the formula, ξ, They are slack variables, and C represents the penalty coefficient, ε represents the deviation, and h i This represents the measured data at each observation point.

[0168] Module M1.3: Solve the nonlinear regression equation of the support vector machine to obtain:

[0169]

[0170] In the formula, α i and Represents the Lagrange multipliers.

[0171] Module M1.4: Based on the solved equations, a support vector machine regression model is constructed using the Laplace kernel function as follows:

[0172] The support vector machine regression model constructed based on the Gaussian kernel function is as follows:

[0173] The support vector machine regression model constructed based on the Sigmoid kernel function is: k3(x i ,x j )=tanh(βx T i x j +θ)

[0174] In the formula, x i and x j Let represent the i-th and j-th interpolation points respectively, δ represent the Laplace kernel parameter and Gaussian kernel parameter, and β and θ represent the Sigmoid kernel parameter.

[0175] Model parameter optimization module: This module performs parameter optimization training on support vector machine regression models based on Gaussian kernel, Laplace kernel, and Sigmoid kernel respectively, to obtain the corresponding optimal parameter combinations. The model parameter optimization module includes the following parameter optimization algorithms, which optimize the regression model parameters for Gaussian kernel support vector machines, Laplace kernel support vector machines, and Sigmoid kernel support vector machines respectively. The formula for calculating the optimal fitness, i.e., the overall loss function, is as follows:

[0176]

[0177] In the formula, h i This represents the measured data at the i-th observation point. This is the interpolated prediction data for the i-th observation point.

[0178] Module M2.1: Initializes the corresponding support vector machine regression model parameters and the corresponding novel parameter optimization algorithm parameters; the novel parameter optimization algorithm parameters include the number of individuals in the population cluster corresponding to each model and the maximum number of iterations of the population; the individuals represent the parameter optimization space.

[0179] Module M2.2: Acquires data and divides the data into regions.

[0180] Module M2.3: Perform interpolation training for each region, calculate the initial value of the optimal fitness, and select the best individual within the population cluster.

[0181] Module M2.4: Performs chaotic optimization on all individuals other than the optimal individual, calculates the current optimal fitness value, and selects the optimal individual within the current population cluster from the new parameter optimization parameters. The chaotic optimization in Module M2.4 is updated using the following formula:

[0182]

[0183]

[0184] In the formula, Let j represent the j-th individual in the k-th generation of the i-th support vector machine regression model. Let j represent the j-th individual in the (k-1)-th generation of the i-th support vector machine regression model. and Let s represent the individuals with the best and worst fitness values ​​in the (k-1)th generation of each population cluster, respectively. k =μ(1-s k-1 ) represents the k-th generation chaotic operator, where 0 <s<1,0<μ≤4,N max This represents the maximum number of iterations for the population.

[0185] Module M2.5: Calculates the number of offspring from the current best individual, and calculates the fitness value of each offspring generated by that individual based on the number of offspring. It then determines whether the fitness value of the offspring generated is better than the current best fitness value. If so, it updates the current best fitness value and the current best individual, and deletes any non-optimal offspring generated by each individual to ensure the overall population size remains unchanged. If not, it does not update. The formula for calculating the number of offspring from the current best individual in module M2.5 is as follows:

[0186]

[0187]

[0188] In the formula, N i,j Let represent the number of offspring of the j-th individual in the i-th support vector machine regression model. and Let these represent the minimum and maximum fitness values ​​in the current k-th generation within the i-th population cluster, respectively. This represents the position fitness value of the j-th individual in the k-th generation of the i-th support vector machine regression model.

[0189] Module M2.6: Divide all individuals in the current population cluster except for the optimal individual into two groups. Perform a tendency operation on the first group and an exploration operation on the second group. Calculate and update the current optimal fitness value and the optimal individual. The tendency operation involves moving towards the global optimum using the following formula:

[0190]

[0191] in, Let j represent the j-th individual in the k-th generation of the i-th support vector machine regression model. This represents the individual in the k-th generation within the cluster that has the optimal fitness value. ε represents a smaller value to avoid invalidation of the denominator. Let be the position fitness value of the j-th individual in the k-th generation of the i-th support vector machine regression model. This represents the optimal fitness value of the i-th support vector machine regression model in the k-th generation of the current population cluster;

[0192] The exploration operation includes performing a random walk using the following formula:

[0193]

[0194] in, Let represent the j-th individual in the k-th generation of the i-th support vector machine regression model, r1 and r2 represent random numbers between 0 and 1, G(1.5, r2) represents Gaussian random numbers, and levy represents the Levy flight update value. It represents the orientation factor of the j-th individual in the k-th generation of the i-th support vector machine regression model.

[0195] Module M2.7: Eliminate the K individuals with the lowest objective function values, regenerate the K individuals in the solution space according to Module M2.1, and then calculate and update the current optimal fitness value and the optimal individual.

[0196] Module M2.8: Determines whether the termination condition has been met. If yes, it terminates the output model and triggers module M2.9; otherwise, it triggers module M2.2.

[0197] Module M2.9: Outputs the Gaussian kernel support vector machine regression model, Laplace kernel support vector machine regression model, and Sigmoid kernel support vector machine regression model corresponding to the optimal parameters.

[0198] The rainfall isosurface generation module updates the support vector machine regression model based on Gaussian kernel, Laplace kernel, and Sigmoid kernel respectively according to the optimized parameter combination to obtain the optimal model. Then, it inputs the data of the points to be interpolated, performs optimal model interpolation, obtains the interpolation results of the corresponding points to be interpolated, and then outputs the interpolation results and isosurface map, while saving the optimal parameters and the error of the corresponding interpolation points.

[0199] Specifically, the rainfall isosurface generation module also includes an incremental sample simulation interpolation and model optimization module and an incremental update optimization training module, described as follows:

[0200] Incremental Sample Simulation Interpolation and Model Optimization Module: When new sample data is added, the input sample dataset is used for simulation interpolation. The results of Gaussian kernel support vector machine regression model, Laplace kernel support vector machine regression model, and Sigmoid kernel support vector machine regression model are output in parallel. The model with the smallest fitness value is selected as the optimal model. Then, it is determined whether the optimal fitness value meets the threshold requirement. If so, the model is selected, and the incremental sample interpolation output is updated directly over time, and the incremental sample interpolation results are saved. If not, the incremental update optimization training module is executed.

[0201] Incremental update optimization training module: This module inputs the current samples in parallel into Gaussian kernel support vector machine interpolation models, Laplace kernel support vector machine interpolation models, and Sigmoid kernel support vector machine interpolation models for model parameter optimization training. According to the present invention, a computer-readable storage medium storing a computer program is provided. When executed by a processor, the computer program implements the steps of the rainfall isosurface generation method based on parameter optimization support vector machine regression.

[0202] According to the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the computer program is executed by the processor, it implements the steps of the rainfall isosurface generation method based on parameter-optimized support vector machine regression.

[0203] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0204] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for generating rainfall isohyet based on parameter optimization support vector machine regression, characterized in that, include: Model building steps: Build support vector machine regression models based on the Laplace kernel, Gaussian kernel, and Sigmoid kernel respectively; Model parameter optimization steps: Perform parameter optimization training on support vector machine regression models based on Gaussian kernel, Laplace kernel and Sigmoid kernel respectively to obtain the corresponding optimal parameter combinations; The steps for generating rainfall isosurfaces are as follows: Based on the optimized parameter combinations, the support vector machine regression models based on Gaussian kernel, Laplace kernel, and Sigmoid kernel are updated respectively to obtain the optimal model. Then, the data of the points to be interpolated are input, the optimal model is used for interpolation, the interpolation results of the corresponding points to be interpolated are obtained, and then the interpolation results and isosurface maps are output. At the same time, the optimal parameters and the errors of the corresponding interpolation points are saved. The model parameter optimization step includes the following parameter optimization algorithms, which optimize the regression model parameters of Gaussian kernel support vector machine, Laplace kernel support vector machine and Sigmoid kernel support vector machine respectively: Step S2.1: Initialize the corresponding support vector machine regression model parameters and the corresponding novel parameter optimization algorithm parameters; the novel parameter optimization algorithm parameters include the number of individuals in the population cluster and the maximum number of iterations for each model corresponding to the novel parameter optimization. The individual representation parameter optimization space; Step S2.2: Acquire data and divide the data into regions; Step S2.3: Perform interpolation training for each region, calculate the initial value of the optimal fitness, and select the best individual within the population cluster; Step S2.4: Perform chaotic optimization on all individuals other than the optimal individual, calculate the current optimal fitness value, and select the optimal individual in the current population cluster from the new parameter optimization parameters; Step S2.5: Calculate the number of offspring in the current best individual, and calculate the fitness value of the offspring generated by each individual based on the number of offspring. Determine whether the fitness value of the offspring generated by each individual is better than the current best fitness value. If so, update the current best fitness value and the current best individual, and delete the non-optimal individuals generated by each individual to ensure that the total population size remains unchanged; if not, do not update. Step S2.6: Divide all individuals in the current population cluster except for the best individual into two groups. Perform a tendency operation on the first group of individuals and an exploration operation on the second group of individuals. Calculate and update the current best fitness value and the best individual. Step S2.7: Eliminate the last ranked objective function value of the individuals and re-generate individuals in the solution space according to step S2.1, and then calculate and update the current optimal fitness value and the optimal individual. Step S2.8: Determine whether the termination condition has been met. If yes, terminate the output model and trigger step S2.9; otherwise, trigger step S2.

2. Step S2.9: Output the Gaussian kernel support vector machine regression model, Laplace kernel support vector machine regression model, and Sigmoid kernel support vector machine regression model corresponding to the optimal parameters.

2. The method of claim 1, wherein, The model construction steps include: Step S1.1: Obtain historical rainfall data as a training sample set, as shown below: wherein, represents the total number of statistical time periods, represents the total number of observation sites, respectively represent the longitude, latitude, elevation, and measured rainfall data within the statistical time period for the i-th observation site, for the j-th observation site. Step S1.2: Based on the training sample set, determine the nonlinear regression equation for the support vector machine as follows: In the formula, express Interpolation prediction data for observation points Represents the kernel function. Indicates the point to be interpolated. , These represent the weight vector and the bias, respectively. The loss function is: In the formula, , They are slack variables, and , , Indicates the penalty coefficient. Indicates deviation, This represents the measured data at each observation point; Step S1.3: Solve the nonlinear regression equation of the support vector machine to obtain: wherein , and denotes the Lagrange multiplier, ; Step S1.4: Constructing the support vector machine regression model based on the Laplacian kernel function through the solved equation is: , The support vector machine regression model is constructed based on a Gaussian kernel function as follows: , The support vector machine regression model is constructed based on a Sigmoid kernel function as follows: , wherein respectively denote the first interpolation point and the second interpolation point, denote a Laplacian kernel parameter and a Gaussian kernel parameter, and denote a Sigmoid kernel parameter.

3. The method of claim 1, wherein the method is characterized by: In step S2.4, the chaotic optimization is updated using the following formula: In the formula, Indicates the first The first support vector machine regression model The first generation individual, Indicates the first The first support vector machine regression model The first generation individual, and Representing the first in each population cluster Individuals with optimal and worst fitness values. Indicates the first The chaotic operator, among which, , , This represents the maximum number of iterations for the population.

4. The method of claim 1, wherein, The formula for calculating the number of offspring in the current optimal individual in step S2.5 is as follows: In the formula, Indicates the first The first support vector machine regression model The number of offspring per individual, and They represent the first The current number within each population cluster The minimum and maximum values ​​of the fitness value. Indicates the first The current i-th support vector machine regression model The first generation Individual location fitness value.

5. The method of claim 1, wherein, The approach operation involves moving closer to the global optimum using the following formula: in, Indicates the first The first support vector machine regression model The first generation individual, Indicates the first term within the cluster The generation has individuals corresponding to the optimal fitness value. , To represent smaller values ​​and avoid invalidation of the denominator, For the first The current i-th support vector machine regression model The first generation Individual position fitness value, Indicates the first The support vector machine regression model is currently the [number]th [number] model within the population cluster. The optimal fitness value; The exploration operation includes performing a random walk using the following formula: in, Indicates the first The first support vector machine regression model The first generation Individual, , express Random numbers between Represents Gaussian random numbers. This indicates the update value of Levi's flight. Indicates the first The first support vector machine regression model The first generation The directional factor of each individual.

6. The method for generating rainfall isosurfaces based on parameter-optimized support vector machine regression according to claim 1, characterized in that, The rainfall isosurface generation process also includes incremental sample simulation interpolation and model optimization steps, and incremental update optimization training steps: Incremental sample simulation interpolation and model optimization steps: When new sample data is added, the input sample dataset is used for simulation interpolation, and the results of Gaussian kernel support vector machine regression model, Laplace kernel support vector machine regression model, and Sigmoid kernel support vector machine regression model are output in parallel. The model with the smallest fitness value is selected as the optimal model. Then, it is determined whether the optimal fitness value meets the threshold requirement. If so, the model is selected, and the incremental sample interpolation output is updated directly over time, and the incremental sample interpolation results are saved. If not, the incremental update optimization training step is executed. Incremental update optimization training steps: Input the current sample into the Gaussian kernel support vector machine interpolation model, the Laplace kernel support vector machine interpolation model, and the Sigmoid kernel support vector machine interpolation model in parallel to optimize the model parameters.

7. A rainfall isohyet generation system based on parameter optimization of support vector machine regression, characterized by, include: Model building module: Build support vector machine regression models based on the Laplace kernel, Gaussian kernel, and Sigmoid kernel respectively; Model parameter optimization module: Perform parameter optimization training on support vector machine regression models based on Gaussian kernel, Laplace kernel and Sigmoid kernel respectively to obtain the corresponding optimal parameter combination; Rainfall isosurface generation module: Based on the optimized parameter combination, update the support vector machine regression model based on Gaussian kernel, Laplace kernel and Sigmoid kernel respectively to obtain the optimal model. Then input the data of the points to be interpolated, perform optimal model interpolation, obtain the interpolation results of the corresponding points to be interpolated, and then output the interpolation results and isosurface map. At the same time, save the optimal parameters and the error of the corresponding interpolation points. The model parameter optimization module employs the following parameter optimization algorithms, which optimize the regression model parameters of Gaussian kernel support vector machine, Laplace kernel support vector machine, and Sigmoid kernel support vector machine, respectively: Step S2.1: Initialize the corresponding support vector machine regression model parameters and the corresponding novel parameter optimization algorithm parameters; the novel parameter optimization algorithm parameters include the number of individuals in the population cluster and the maximum number of iterations for each model corresponding to the novel parameter optimization. The individual representation parameter optimization space; Step S2.2: Acquire data and divide the data into regions; Step S2.3: Perform interpolation training for each region, calculate the initial value of the optimal fitness, and select the best individual within the population cluster; Step S2.4: Perform chaotic optimization on all individuals other than the optimal individual, calculate the current optimal fitness value, and select the optimal individual in the current population cluster from the new parameter optimization parameters; Step S2.5: Calculate the number of offspring in the current best individual, and calculate the fitness value of the offspring generated by each individual based on the number of offspring. Determine whether the fitness value of the offspring generated by each individual is better than the current best fitness value. If so, update the current best fitness value and the current best individual, and delete the non-optimal individuals generated by each individual to ensure that the total population size remains unchanged; if not, do not update. Step S2.6: Divide all individuals in the current population cluster except for the best individual into two groups. Perform a tendency operation on the first group of individuals and an exploration operation on the second group of individuals. Calculate and update the current best fitness value and the best individual. Step S2.7: Eliminate the last ranked objective function value of the individuals and re-generate the individuals in the solution space according to step S2.1, and then calculate and update the current optimal fitness value and the optimal individual. Step S2.8: Determine whether the termination condition has been met. If yes, terminate the output model and trigger step S2.9; otherwise, trigger step S2.

2. Step S2.9: Output the Gaussian kernel support vector machine regression model, Laplace kernel support vector machine regression model, and Sigmoid kernel support vector machine regression model corresponding to the optimal parameters.

8. A computer readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the rainfall isosurface generation method based on parameter optimization support vector machine regression as described in any one of claims 1 to 6.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the rainfall isosurface generation method based on parameter optimization support vector machine regression as described in any one of claims 1 to 6.

Images