Permafrost upper limit prediction method and device based on modal decomposition, terminal and storage medium

Abstract

The invention provides a permafrost upper limit prediction method and device based on modal decomposition, a terminal and a storage medium, and belongs to the technical field of permafrost monitoring. The method comprises the following steps: acquiring frozen soil upper limit influence factor data of a to-be-predicted region; performing variational mode decomposition on each piece of influence factor data in the frozen soil upper limit influence factor data to obtain an intrinsic component of the frozen soil upper limit influence factor data; calculating a correlation coefficient of the intrinsic component and the frozen soil upper limit, and determining intrinsic component data with high correlation with the frozen soil upper limit according to the correlation coefficient; inputting the intrinsic component data into a pre-trained frozen soil upper limit fusion prediction model to obtain a frozen soil upper limit prediction result; wherein the frozen soil upper limit fusion prediction model is obtained by training based on a pre-constructed data set, the data set comprises intrinsic component data highly related to a historical frozen soil upper limit and the historical frozen soil upper limit, and the frozen soil upper limit fusion prediction model comprises a gradient lifting sub-model and a self-adaptive sub-model. The accuracy of frozen soil upper limit prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of frozen soil monitoring, and in particular to a method, device, terminal and storage medium for predicting the upper limit of permafrost based on modal decomposition. Background Art

[0002] Permafrost is a kind of soil with special properties. The upper limit of permafrost refers to the upper boundary of the permafrost layer. The change of the upper limit of permafrost is affected by many factors, including climate change, surface cover, human activities, etc. These factors lead to unstable changes in the upper limit of permafrost. The change of the upper limit of permafrost directly reflects the stability and dynamic characteristics of the permafrost layer. Permafrost will experience phenomena such as frost heave and thaw settlement under different climate and seasonal changes, especially the decline of the upper limit of permafrost, which will pose a serious deformation threat to the roadbed, thereby affecting the stability and safety of the road. Therefore, accurate prediction of the upper limit of permafrost is extremely important for the management and maintenance of roadbed diseases.

[0003] Traditional permafrost upper limit prediction methods usually rely on long-term monitoring of meteorological data, geological data and the physical properties of frozen soil itself. These data are directly used to establish prediction models, based on linear assumptions or simplified physical processes in order to achieve accurate predictions of changes in the upper limit of permafrost. However, due to the complexity and nonlinear characteristics of the permafrost system itself and the difficulty in grasping the temporal changes of data, traditional methods have many limitations. For example, the degree of influence of different data on the changes in the upper limit of permafrost varies greatly, and a large amount of data with low correlation participates in model training, reducing the accuracy of prediction. The prediction methods of the prior art are limited by the complexity of the data, temporal changes and nonlinear relationships, resulting in low accuracy of frozen soil prediction. Summary of the invention

[0004] The embodiments of the present invention provide a method, device, terminal and storage medium for predicting the upper limit of permafrost based on modal decomposition to solve the problem of low accuracy of the predicted upper limit of permafrost.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting an upper limit of permafrost based on modal decomposition, comprising: Obtain data on factors affecting the upper limit of frozen soil in the area to be predicted; the data on factors affecting the upper limit of frozen soil include temperature, humidity, rainfall and wind speed; The data of factors influencing the upper limit of frozen soil are preprocessed, and the data of each factor influencing the upper limit of frozen soil are respectively subjected to variational mode decomposition to obtain the intrinsic components of the data of factors influencing the upper limit of frozen soil; wherein the intrinsic components include periodic terms, trend terms and residual terms; Calculate the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil, and determine the intrinsic component data with high correlation with the upper limit of frozen soil according to the correlation coefficient; The intrinsic component data is input into a pre-trained permafrost upper limit fusion prediction model to obtain the permafrost upper limit prediction result; wherein, the permafrost upper limit fusion prediction model is trained based on a pre-constructed data set, the data set includes intrinsic component data with high correlation with the historical permafrost upper limit and the historical permafrost upper limit, and the permafrost upper limit fusion prediction model includes a gradient boosting sub-model and an adaptive sub-model.

[0006] In a possible implementation, the gradient boosting sub-model is a gradient boosting tree model, the adaptive sub-model is an adaptive boosting model, and the training process of the frozen soil upper limit fusion prediction model includes: The error rate of the gradient boosting tree model is calculated according to the intrinsic component data, and the weight of the gradient boosting tree model is calculated according to the error rate of the gradient boosting tree model; the weight of the data set is calculated according to the weights of the adaptive enhancement model and the gradient boosting tree model, the data set is updated according to the weight of the data set, and a new gradient boosting tree model is obtained by training according to the updated data set. When the number of gradient boosting tree models reaches a preset number, all trained gradient boosting tree models are combined to obtain a permafrost upper limit fusion prediction model.

[0007] In a possible implementation, the error rate of the gradient boosting tree model is calculated according to the eigencomponent data, and the weight of the gradient boosting tree model is calculated according to the error rate of the gradient boosting tree model, including: The intrinsic component data with high correlation with the historical upper limit of frozen soil are input into the gradient boosting tree model to obtain the predicted upper limit of frozen soil; The error rate of the gradient boosting tree model was calculated based on the predicted upper and historical upper limits of frozen soil; Combined with the first preset formula, the weight of the gradient boosting tree model in the permafrost upper limit fusion prediction model is calculated according to the error rate. The first preset formula is: =

[0008] In the formula, is the weight of the gradient boosted tree model, is the error rate of the gradient boosted tree model.

[0009] In a possible implementation, the weight of the data set is calculated according to the weights of the adaptive boosting model and the gradient boosting tree model, including: Combined with the second preset formula, the weight of the data set is calculated according to the weight of the gradient boosting tree model, the predicted upper limit of frozen soil, the upper limit of historical frozen soil and the weight of the previous data set. The second preset formula is:

[0010] In the formula, It is t The weight of the dataset at the iteration, It is i The weights of the gradient boosted tree model, It is a dataset i The historical upper limit of permafrost, It is t Gradient boosted tree model on the dataset i The predicted upper limit of permafrost is the weight normalization numerator, and the formula is .

[0011] In one possible implementation, all trained gradient boosting tree models are combined to obtain a permafrost upper limit fusion prediction model, including: All trained gradient boosting tree models and the weights corresponding to each gradient boosting tree model are multiplied and then summed up to obtain the permafrost upper limit fusion prediction model.

[0012] In a possible implementation, the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil is calculated, including: Combined with the third preset formula, the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil is calculated. The third preset formula is:

[0013] In the formula, is the Spearman rank correlation coefficient, is the rank difference of each pair of observations in the two variables, is the sample size.

[0014] In a possible implementation, each influencing factor data in the preprocessed permafrost upper limit influencing factor data is subjected to variational mode decomposition, including: Combined with the fourth preset formula, the preprocessed frozen soil upper limit influencing factor data are subjected to variational mode decomposition respectively. The fourth preset formula is:

[0015] In the formula, is the Wiener filter of the residual at the current moment, is the number of iterations, ^ is the Fourier transform symbol, is the original signal at frequency The Fourier transform value at It is The modal signal at the frequency The estimated value at is the Lagrange multiplication operator at frequency The value at is the penalty factor, is the frequency, is the center frequency of the component.

[0016] In a second aspect, an embodiment of the present invention provides a permafrost upper limit prediction device based on modal decomposition, comprising: A data acquisition module is used to acquire data on factors affecting the upper limit of frozen soil in the area to be predicted; wherein the data on factors affecting the upper limit of frozen soil include temperature, humidity, rainfall and wind speed; The intrinsic component acquisition module is used to preprocess the data of factors affecting the upper limit of frozen soil, and perform variational mode decomposition on the data of each factor affecting the upper limit of frozen soil after preprocessing, so as to obtain the intrinsic components of the data of factors affecting the upper limit of frozen soil; wherein the intrinsic components include periodic terms, trend terms and residual terms; An intrinsic component data determination module is used to calculate the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil, and determine the intrinsic component data with high correlation with the upper limit of frozen soil according to the correlation coefficient; The permafrost upper limit prediction module is used to input the intrinsic component data into the pre-trained permafrost upper limit fusion prediction model to obtain the permafrost upper limit prediction result; wherein, the permafrost upper limit fusion prediction model is trained based on a pre-constructed data set, and the data set includes the intrinsic component data with high correlation with the historical permafrost upper limit and the historical permafrost upper limit, and the permafrost upper limit fusion prediction model includes a gradient boosting sub-model and an adaptive sub-model.

[0017] In a third aspect, an embodiment of the present invention provides a terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the first aspect or any possible implementation method of the first aspect are implemented.

[0018] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method in the first aspect or any possible implementation of the first aspect.

[0019] The embodiment of the present invention provides a method, device, terminal and storage medium for predicting the upper limit of permafrost based on modal decomposition. The intrinsic components of the influencing factor data of the upper limit of permafrost are subjected to correlation analysis, and the intrinsic components with high correlation with the upper limit of permafrost are selected as the input features of permafrost prediction. The selected features are input into the fusion prediction model of the upper limit of permafrost to obtain the prediction result of the upper limit of permafrost. The data with high correlation with the upper limit of permafrost are used as the input features for predicting the upper limit of permafrost to improve the accuracy of prediction. The fusion prediction model of the upper limit of permafrost includes a gradient boosting sub-model and an adaptive sub-model. The gradient boosting sub-model optimizes the prediction performance of the model by gradually fitting the residual based on the principle of gradient boosting. The adaptive sub-model adaptively optimizes the performance of the model by dynamically adjusting the sample weights, so that the model can gradually improve the classification ability of complex data. The final fusion prediction model of the upper limit of permafrost not only utilizes the advantages of each model, but also reduces the error of a single model, which can significantly improve the generalization ability and prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 is a flow chart for implementing a method for predicting the upper limit of permafrost based on modal decomposition provided by an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a permafrost upper limit prediction device based on modal decomposition provided by an embodiment of the present invention; Figure 3 is a schematic diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0023] In order to make the purpose, technical solutions and advantages of the present invention more clear, specific embodiments will be described below in conjunction with the accompanying drawings.

[0024] Figure 1The following is a flowchart for implementing the upper limit prediction of permafrost based on modal decomposition provided by the embodiment of the present invention. Figure 1 As shown, including: Step 101: Obtain data on factors affecting the upper limit of frozen soil in the area to be predicted; wherein the data on factors affecting the upper limit of frozen soil include temperature, humidity, rainfall and wind speed.

[0025] In this embodiment, in the process of predicting the upper limit of frozen soil, it is necessary to collect data on various influencing factors related to the upper limit of frozen soil in the area to be predicted, such as temperature, humidity, rainfall and wind speed, wherein changes in temperature can affect the freezing and thawing process of frozen soil. Humidity refers to soil moisture, which can affect the heat conduction and water migration of frozen soil. Precipitation can affect the moisture content of the soil, and thus affect the thermal state and stability of frozen soil. Wind speed can affect the heat exchange and evaporation process of the surface, and indirectly affect the temperature conditions of frozen soil. These factors are key variables that affect the changes in the upper limit of frozen soil. By obtaining these data, input can be provided for the subsequent fusion prediction model of the upper limit of frozen soil, helping to more accurately predict the changes in the upper limit of frozen soil.

[0026] Step 102: preprocess the data of factors influencing the upper limit of frozen soil, and perform variational mode decomposition on the data of each factor influencing the upper limit of frozen soil after preprocessing, so as to obtain the intrinsic components of the data of factors influencing the upper limit of frozen soil; wherein the intrinsic components include periodic terms, trend terms and residual terms.

[0027] In this embodiment, variational mode decomposition is a new, non-recursive, fully intrinsic and adaptive signal processing tool that can transform the signal Decomposed into sub-signals or intrinsic mode functions (IMFs). These IMFs are spectral components of the original signal, have limited bandwidth, and are compact around the central pulsation. Each IMF represents the characteristics of the original signal in a different frequency range, and by combining all IMFs, the original signal can be reconstructed. In order to detect the bandwidth and center frequency of each IMF, variational mode decomposition uses the optimal results of the evolutionary variational problem as a theoretical solution and iteratively maintains a constant optimization strategy.

[0028] Specifically, for the upper limit of permafrost, time series signals such as temperature, humidity, rainfall, wind speed and heat flux It can be seen as the superposition of the overall trend and local fluctuations. From the overall trend, It has a periodicity of days, and the mutual influence of various factors may lead to obvious local fluctuations. The non-recursive variational mode decomposition method can effectively overcome the endpoint effect and mode aliasing when facing time series, and its robustness and adaptability are better. Using variational mode decomposition to decompose the influencing factors can obtain three IMFs, namely the periodic term, trend term and residual term, which can better highlight its volatility, periodicity, trend and other characteristics.

[0029] Variational mode decomposition obtains the frequency and bandwidth of each IMF by iteratively solving the variational optimization problem, thereby achieving the effect of adaptive decomposition and noise reduction of the original signal. The objective function of the decomposition method is as follows:

[0030] In the objective function For the k Component at time t The sample at For the k Component at time t The center frequency is For the moment t The differential of is the impulse function per unit time, Represents the L2 norm. The bandwidth constraint problem corresponding to each mode is transformed into an unconstrained problem by introducing the augmented Lagrangian function. The solution is as follows:

[0031] In the formula, is the augmented Lagrangian function, is the Lagrange multiplication operator, is the penalty factor, Represents the inner product.

[0032] In a possible implementation, the preprocessed frozen soil upper limit influencing factor data are subjected to variational mode decomposition respectively. The specific processing is as follows: in combination with the fourth preset formula, the preprocessed frozen soil upper limit influencing factor data are subjected to variational mode decomposition respectively. The fourth preset formula is:

[0033] In the formula, is the Wiener filter of the residual at the current moment, is the number of iterations, ^ is the Fourier transform symbol, is the original signal at frequency The Fourier transform value at It is The modal signal at the frequency The estimated value at is the Lagrange multiplication operator at frequency The value at is the penalty factor, is the frequency, is the center frequency of the component.

[0034] The calculation formula of the center frequency is:

[0035] In the formula, is the minimum value of the center frequency of the modal component.

[0036] In this example, the original data of each influencing factor of the permafrost upper limit are decomposed into different components , and calculate the center frequencies corresponding to different components .

[0037] In one possible implementation, the preprocessing is specifically performed as follows: cleaning the data of factors affecting the upper limit of frozen soil; performing outlier detection on the cleaned data of factors affecting the upper limit of frozen soil, and filling in the missing or outlier values ​​in the cleaned data of factors affecting the upper limit of frozen soil; resampling the filled data of factors affecting the upper limit of frozen soil; normalizing the resampled data of factors affecting the upper limit of frozen soil, and obtaining the preprocessed data of factors affecting the upper limit of frozen soil.

[0038] In this embodiment, the meteorological data of the area, including temperature, humidity, rainfall, wind speed and other information, is obtained through the Meteorological Bureau, and the ground temperature data collected by sensors arranged at different depths of the railway roadbed is received. The upper limit of frozen soil and heat flux data are calculated, and the above-obtained data are sorted and imported into the original database. The raw data obtained in this way often cannot be used directly, and the data needs to be preprocessed, usually with missing value processing, outlier detection and processing, data cleaning, and data normalization processing, so that the data is on the same scale to improve the efficiency of model training.

[0039] Specifically, the data collected by the monitoring sensors may be affected by environmental interference and equipment problems, resulting in abnormal isolated data points and noise data. These abnormal data will affect the accuracy and real-time performance of the upper limit prediction and stability assessment of frozen soil, so data cleaning is required. In order to clean the data collected by the monitoring sensors along the railway, we chose to use Data denoising is performed based on the principle to eliminate abnormal data and noise and ensure data quality and the reliability of analysis results.

[0040] Secondly, during the data processing, the original data may contain missing or abnormal values, and data filling is required to ensure the continuity of the data. Especially for data with strong continuity such as ground temperature and meteorological data, the adjacent value filling method can be used to fill the missing values ​​during processing. Since these data do not change much in a short period of time, the adjacent value filling method can easily complete the data filling operation while maintaining the accuracy of the filling results and avoiding the introduction of large errors.

[0041] Thirdly, since the sampling frequency of ground temperature data is twice a month, which is different from the frequency of air temperature data, the data needs to be resampled and frequency converted. One method is to convert the air temperature data to a frequency of twice a month by calculating the average of the air temperature data in each time period. This can keep the frequency of ground temperature data consistent with that of air temperature data for subsequent analysis and comparison.

[0042] Then, the present invention considers using the gradient descent method in the model training stage. Since the ranges of values ​​of various continuous data features (such as temperature, humidity, heat flux, and upper limit of frozen soil) vary greatly, in order to speed up the convergence of the model, it is necessary to normalize these continuous feature data. Through normalization, the ranges of values ​​of different feature data can be mapped to similar scales, which is conducive to faster learning and convergence of the model and improves training efficiency and accuracy.

[0043] Step 103: Calculate the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil, and determine the intrinsic component data with high correlation with the upper limit of frozen soil according to the correlation coefficient.

[0044] In a possible implementation, the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil is calculated. The specific processing is as follows: Combined with the third preset formula, the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil is calculated. The third preset formula is:

[0045] In the formula, is the Spearman rank correlation coefficient, is the rank difference of each pair of observations in the two variables, is the sample size.

[0046] Accordingly, the upper limit of permafrost is inferred based on the distribution and conduction characteristics of ground temperature. The temperature at the upper limit of permafrost is usually close to 0°C, and the ground temperature increases with depth below this depth. The ground temperature data at different depths in the relevant area are collected by the data collector, and the curve of ground temperature change with depth is drawn. The inflection point in the ground temperature curve where the temperature is close to 0°C and increases with depth is found. This inflection point is the upper limit of permafrost.

[0047] In this embodiment, the decomposed data set is processed centrally, and the Spearman rank correlation coefficient is used to calculate the correlation coefficient between the upper limit of frozen soil and the different intrinsic components obtained after the decomposition of each influencing factor, and the intrinsic components of the influencing factors with a higher than preset correlation are selected as the relevant data to select the key indicators that affect the change of the upper limit of frozen soil and improve the prediction accuracy of the upper limit of permafrost. The Spearman rank correlation coefficient is a non-parametric statistic used to measure the correlation between two variables. It calculates the correlation by comparing the ranks of the variables instead of the original values. Therefore, it is also effective for nonlinear relationships and is conducive to improving the accuracy of model prediction.

[0048] Step 104: Input the intrinsic component data into a pre-trained permafrost upper limit fusion prediction model to obtain a permafrost upper limit prediction result; wherein the permafrost upper limit fusion prediction model is trained based on a pre-constructed data set, the data set includes intrinsic component data with a high correlation with the historical permafrost upper limit and the historical permafrost upper limit, and the permafrost upper limit fusion prediction model includes a gradient boosting sub-model and an adaptive sub-model.

[0049] In a possible implementation, the gradient boosting sub-model is a gradient boosting tree model, the adaptive sub-model is an adaptive boosting model, and the training process of the permafrost upper limit fusion prediction model includes: calculating the error rate of the gradient boosting tree model based on the intrinsic component data, and calculating the weight of the gradient boosting tree model based on the error rate of the gradient boosting tree model; calculating the weight of the data set based on the weights of the adaptive boosting model and the gradient boosting tree model, updating the data set based on the weight of the data set, and obtaining a new gradient boosting tree model through training based on the updated data set. When the number of gradient boosting tree models reaches a preset number, all trained gradient boosting tree models are combined to obtain the permafrost upper limit fusion prediction model.

[0050] In this embodiment, the gradient boosting tree model is an iterative decision tree algorithm, which corrects the residual of the previous round of prediction by continuously adding new decision trees, thereby improving the prediction ability of the model. The present invention uses the gradient boosting tree model to predict frozen soil. The gradient boosting tree uses the CART regression tree as a weak classifier. Each new model established is based on the negative gradient of the loss function of the previous model, which continuously reduces the model residual in successive iterations, and finally forms an integrated model with strong prediction ability. The nonlinear relationship and complex relationship in the data are processed by the gradient boosting tree model. The calculation steps are as follows: First, determine the training sample set.

[0051]

[0052] in, For the n The characteristic value of a sample is m dimensional vector, For the n The output value of the sample, .

[0053] Second, initialize a weak learner.

[0054]

[0055] in, L is the loss function, , The function returns the value that satisfies the loss function. L The parameter c corresponding to the minimum value.

[0056] Third, the negative gradient value of the current model loss function (i.e., the estimated value of the residual) is as follows.

[0057]

[0058] Fourth, As the training data for the next tree, the estimated value of the fitting residual is used to obtain a new regression tree. The area of ​​each leaf node in the regression tree is ( j=1,2,…,J, in J is the number of leaf nodes).

[0059] Fifth, the best gradient fitting value of each node is as follows.

[0060]

[0061] Sixth, based on the above result values, the strong learner is updated.

[0062]

[0063] in, .

[0064] Seventh, after the iterative process of the lifting method, the final model is output.

[0065]

[0066] The adaptive boosting model is an iterative algorithm. Its core idea is to train different weak classifiers for the same training set, and then combine these weak classifiers to form a strong classifier that can handle classification and regression problems. The adaptive boosting model algorithm process is as follows: Suppose there is a set of sample training sets ,in, , represents the category label of the training sample, The value range is , the total number of base models trained is . No. The weight of each sample in the training set of the base model is , Representative The base model is in the training set The sample weights.

[0067] First, initialize the weight distribution in the training dataset ,in , when initialized, all data sample weights are , therefore, the initial weight of the training sample set is: .

[0068] Secondly, the training data set The weight distribution is calculated as The base model obtained from the training samples The error rate is: ,in is the error rate, is the indicator function. The error rates involved in the present invention all refer to the error rates of the gradient boosting tree model.

[0069] In a possible implementation, the error rate of the gradient boosting tree model is calculated according to the intrinsic component data, and the weight of the gradient boosting tree model is calculated according to the error rate of the gradient boosting tree model. The specific processing is as follows: the intrinsic component data with high correlation with the historical frozen soil upper limit is input into the gradient boosting tree model to obtain the predicted frozen soil upper limit; the error rate of the gradient boosting tree model is calculated according to the predicted frozen soil upper limit and the historical frozen soil upper limit; combined with the first preset formula, the weight of the gradient boosting tree model in the frozen soil upper limit fusion prediction model is calculated according to the error rate. The first preset formula is: =

[0070] In the formula, is the weight of the gradient boosted tree model, is the error rate of the gradient boosted tree model.

[0071] Specifically, the error rate is measured by indicators such as mean square error (MSE) and mean absolute error (MAE).

[0072] In this embodiment, the weight of the gradient boosting tree model in the frozen soil upper limit fusion prediction model is calculated according to the error rate, and the weight of each gradient boosting tree model in the final frozen soil upper limit fusion prediction model is determined. The error rate reflects the accuracy of the model prediction.

[0073] Then, update the weight distribution of the training sample data .

[0074] In a possible implementation, the weight of the data set is calculated according to the weights of the adaptive boosting model and the gradient boosting tree model. The specific processing is as follows: in combination with the second preset formula, the weight of the data set is calculated according to the weight of the gradient boosting tree model, the predicted upper limit of frozen soil, the historical upper limit of frozen soil and the weight of the previous data set. The second preset formula is:

[0075] In the formula, It is t The weight of the dataset at the iteration, It is i The weights of the gradient boosted tree model, It is a dataset i The historical upper limit of permafrost, It is t Gradient boosted tree model on the dataset i The predicted upper limit of permafrost is the weight normalization numerator, and the formula is .

[0076] In this embodiment, the weight of the data set is updated according to the prediction results of the weak learner. The weight of the samples that are misclassified is increased, while the weight of the samples that are correctly classified is reduced, so that in the next round of iteration, the model pays more attention to those samples that are difficult to classify. The adaptive enhancement model is used to further optimize the permafrost upper limit fusion prediction model to improve its prediction accuracy and adaptability under different environmental conditions.

[0077] In one possible implementation, all trained gradient boosting tree models are combined to obtain a permafrost upper limit fusion prediction model. The specific process is as follows: all trained gradient boosting tree models and the weights corresponding to each gradient boosting tree model are multiplied and then summed to obtain the permafrost upper limit fusion prediction model.

[0078] In this embodiment, the method for constructing the permafrost upper limit fusion prediction model enables the model to pay attention to the misclassified samples while making full use of the information of each weak learner, thereby improving the overall classification accuracy. The final permafrost upper limit fusion prediction model is formed by integration, and the formula is: .

[0079] In a possible implementation, the preset number of models is determined according to a Bayesian optimization function.

[0080] In this embodiment, Bayesian optimization can automatically find the optimal preset number within a preset search range so that the prediction error of the model on the test set is minimized.

[0081] In one possible implementation, during the data training and prediction process, a 5-fold cross-validation method is used for small sample data, and the training data is divided into five equal parts, four of which are training sets and the other is a validation set, and the combination is arranged accordingly to achieve the effect of improving the accuracy of the prediction model. In addition, grid search, random search or Bayesian optimization are also used for hyperparameter optimization in parameter optimization. The optimal parameters are shown in Table 1.

[0082] Table 1 Model parameter configuration

[0083] First, the final adaptive boosting model is retrained using the best hyperparameters found. The dataset is divided into training and test sets again, and then the final gradient boosting tree model and adaptive boosting model are created and trained. The trained permafrost upper limit fusion prediction model is used to predict the training and test sets, and the prediction results and true values ​​are inversely transformed to obtain the scale of the original data. The mean square error (MSE) and determination coefficient (R²) of the training and test sets are calculated to evaluate the performance of the permafrost upper limit fusion prediction model. .

[0084] The embodiment of the present invention performs correlation analysis on each intrinsic component of the influencing factor data of the upper limit of frozen soil, selects the intrinsic component with high correlation with the upper limit of frozen soil as the input feature of frozen soil prediction, inputs the selected feature into the fusion prediction model of the upper limit of frozen soil, obtains the prediction result of the upper limit of frozen soil, and uses the data with high correlation with the upper limit of frozen soil as the input feature of predicting the upper limit of frozen soil to improve the accuracy of prediction. The error rate of the gradient boosting tree model is calculated according to the intrinsic component data, and the weight of the gradient boosting tree model is calculated according to the error rate of the gradient boosting tree model; the weight of the data set is calculated according to the weight of the adaptive enhancement model and the gradient boosting tree model, the data set is updated according to the weight of the data set, and a new gradient boosting tree model is obtained by training the updated data set. When the number of gradient boosting tree models reaches a preset number, all the trained gradient boosting tree models are combined to obtain the fusion prediction model of the upper limit of frozen soil. By adjusting the weight of the data set, the model will pay more attention to the data that is difficult to classify, so that the model can reduce errors in each iteration, thereby gradually improving the accuracy of the overall prediction. The final fusion prediction model of the upper limit of frozen soil not only utilizes the advantages of each model, but also reduces the error of a single model, which can significantly improve the generalization ability and prediction accuracy of the model.

[0085] In other possible ways, variational mode decomposition, gradient boosting tree model, adaptive boosting model and Spearman rank correlation coefficient can be implemented in other ways. This application mainly uses variational mode decomposition, gradient boosting tree model, adaptive boosting model and Spearman rank correlation coefficient four-stage processing as an example for explanation.

[0086] For example, the gradient boosting tree model in this application can be replaced by a random forest model, a support vector machine model, XGBoost or LightGBM. Random forest (RF) is used as a base learner to construct multiple decision trees by randomly selecting samples and features. AdaBoost improves the performance of weak learners by iteratively adjusting sample weights. In each iteration, AdaBoost adjusts the sample weights according to the prediction error of the current base learner, so that subsequent base learners pay more attention to misclassified samples. In the RF-AdaBoost model, a random forest base learner is trained in each iteration and its weight is calculated. Finally, the prediction results of all base learners are weighted averaged. The diversity of random forests combined with the adaptive adjustment of AdaBoost can better handle the complex feature relationships in frozen soil prediction.

[0087] Support vector machine (SVM) is used as the base learner, and the appropriate kernel function (such as RBF kernel) is selected and the parameters (such as penalty parameter C and kernel parameter γ) are adjusted. In the process of training the SVM model, the support vector is found by optimizing the problem. AdaBoost adjusts the sample weights in each iteration to make SVM pay more attention to samples that are difficult to classify. Finally, a strong learner is generated through weighted combination to obtain the final prediction value. The combination of SVM's kernel technique and AdaBoost's adaptive enhancement can effectively improve the model's ability to capture complex feature relationships.

[0088] XGBoost is used as a base learner, and appropriate regularization parameters and kernel parameters (such as RBF kernel) are selected during the training process. The model is trained by optimizing the first and second order derivatives of the loss function. AdaBoost adjusts the model weights according to the error rate in each iteration, and generates the final strong learner through weighted combination. The combination of XGBoost's high efficiency and AdaBoost's adaptability can better cope with the complex data structure in frozen soil prediction.

[0089] LightGBM is used as a base learner, and appropriate regularization parameters and kernel parameters are selected during the training process. The model is trained by optimizing the first-order and second-order derivatives of the loss function. AdaBoost adjusts the model weights according to the error rate in each iteration, and generates the final strong learner through weighted combination. The combination of LightGBM's efficient training capability and AdaBoost's adaptive enhancement capability can significantly improve the model's training speed and prediction accuracy, which is especially suitable for processing large-scale time series data involved in frozen soil prediction.

[0090] For example, the adaptive enhancement model in the present application can be replaced by a random forest model, a support vector machine model, or a neural network (such as a multi-layer perceptron MLP) model.

[0091] Random forest is an ensemble learning method based on decision trees. It generates multiple decision trees by introducing sample sampling and feature sampling during the training process, and makes predictions by voting or averaging. Random forest improves generalization ability by constructing multiple independent decision trees. Each tree is trained on randomly selected samples and feature subsets. It does not need to adjust sample weights like AdaBoost, nor does it need to focus on residuals like GBT. Instead, it uses randomness to reduce overfitting and improve the robustness of the model. When training a random forest model, it is necessary to determine hyperparameters (such as the number of trees, maximum depth, feature sampling ratio, etc.) and use optimization methods (such as grid search or Bayesian optimization) for tuning. In the model combination stage, the prediction results of GBT and random forest can be fused through weighted averaging or voting mechanisms.

[0092] SVM is a classification and regression model based on the principle of maximizing the margin. It maps data into a high-dimensional space to find the optimal segmentation hyperplane. GBT training remains unchanged, and the output features of GBT are used to train SVM. When training the SVM model, appropriate kernel function parameters (such as the γ value of the RBF kernel) and penalty parameter C are selected, and parameter optimization methods (such as cross-validation) are used. When assigning weights, the weights of the GBT and SVM models can be determined by Bayesian optimization, and the prediction results of GBT and SVM are combined using weighted average to output the final prediction results.

[0093] Neural networks are usually trained by back propagation and gradient descent, so the weight update mechanism during model training needs to be adjusted. In the data preprocessing stage, the data needs to be normalized or standardized to adapt to the data scale requirements of the neural network. The training process of GBT remains unchanged and is used to extract preliminary features or make preliminary predictions. For neural networks, it is necessary to clarify their structure, such as the number of layers, the number of neurons in each layer, the activation function, etc., determine the choice of loss function (such as mean square error or cross entropy loss), optimizer (such as Adam or SGD), and the number of training iterations. Grid search, cross validation, or Bayesian optimization can be used to determine the hyperparameters of the neural network. When allocating model weights, the weights of GBT and neural network are determined by cross validation or Bayesian optimization, and the prediction results of GBT and neural network are combined using weighted average to output the final prediction results.

[0094] The present invention uses variational mode decomposition to decompose the data of factors affecting the upper limit of frozen soil (air temperature, humidity, wind speed, rainfall, near-surface temperature and heat flux), performs Spearman rank correlation analysis on the decomposed data, selects data with correlation greater than the set value as feature items, and inputs the selected feature data into the upper limit fusion prediction model of frozen soil combined with the gradient boosting tree model and the adaptive boosting model. The gradient boosting tree model is used as a base model to correct the errors of the previous model by gradually adding new tree models, while the adaptive boosting model improves the accuracy of prediction by combining multiple weak classifiers. The advantage of the gradient boosting tree model is that it can flexibly handle various types of data, is suitable for linear and nonlinear regression problems, and has a high prediction accuracy under relatively short parameter adjustment time. However, the gradient boosting tree model is sensitive to outliers, and due to the serial dependence of weak classifiers, it is difficult to train data in parallel. The adaptive boosting model mainly focuses on reducing deviations, and can build a strong integrated model based on a learner with relatively weak generalization performance. Its advantage is that it can improve the generalization ability of the model, especially in the presence of noise, but its disadvantage is that it is more sensitive to outliers. Preprocessing data by variational mode decomposition can reduce the overlap between features and improve the separability of features. The combination of the gradient boosting tree model and the adaptive boosting model can improve the robustness to outliers while maintaining the generalization ability of the model, and ultimately output a better prediction result.

[0095] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0096] The following is an embodiment of the device of the present invention. For details not described in detail therein, reference may be made to the corresponding method embodiment described above.

[0097] Figure 2 The schematic diagram of the structure of the permafrost upper limit prediction device based on modal decomposition provided by an embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows: like Figure 2 As shown, the permafrost upper limit prediction device 2 based on modal decomposition includes: The data acquisition module 21 is used to acquire the data of factors affecting the upper limit of frozen soil in the area to be predicted; wherein the data of factors affecting the upper limit of frozen soil include temperature, humidity, rainfall and wind speed; The intrinsic component acquisition module 22 is used to pre-process the frozen soil upper limit influencing factor data, and perform variational mode decomposition on each influencing factor data in the pre-processed frozen soil upper limit influencing factor data to obtain the intrinsic components of the frozen soil upper limit influencing factor data; wherein the intrinsic components include periodic terms, trend terms and residual terms; An intrinsic component data determination module 23 is used to calculate the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil, and determine the intrinsic component data with high correlation with the upper limit of frozen soil according to the correlation coefficient; The permafrost upper limit prediction module 24 is used to input the intrinsic component data into a pre-trained permafrost upper limit fusion prediction model to obtain a permafrost upper limit prediction result; wherein the permafrost upper limit fusion prediction model is trained based on a pre-constructed data set, and the data set includes intrinsic component data with a high correlation with the historical permafrost upper limit and the historical permafrost upper limit, and the permafrost upper limit fusion prediction model includes a gradient boosting sub-model and an adaptive sub-model.

[0098] In a possible implementation, the frozen soil upper limit prediction module 24 may be used to: The error rate of the gradient boosting tree model is calculated based on the eigencomponent data, and the weight of the gradient boosting tree model is calculated based on the error rate of the gradient boosting tree model, including: The intrinsic component data with high correlation with the historical upper limit of frozen soil are input into the gradient boosting tree model to obtain the predicted upper limit of frozen soil; The error rate of the gradient boosting tree model was calculated based on the predicted upper and historical upper limits of frozen soil; Combined with the first preset formula, the weight of the gradient boosting tree model in the permafrost upper limit fusion prediction model is calculated according to the error rate. The first preset formula is: =

[0099] In the formula, is the weight of the gradient boosted tree model, is the error rate of the gradient boosted tree model.

[0100] In a possible implementation, the frozen soil upper limit prediction module 24 may be used to: The weight of the data set is calculated based on the weights of the adaptive boosting model and the gradient boosting tree model, including: Combined with the second preset formula, the weight of the data set is calculated according to the weight of the gradient boosting tree model, the predicted upper limit of frozen soil, the upper limit of historical frozen soil and the weight of the previous data set. The second preset formula is:

[0101] In the formula, It is t The weight of the dataset at the iteration, It is i The weights of the gradient boosted tree model, It is a dataset i The historical upper limit of permafrost, It is tGradient boosted tree model on the dataset i The predicted upper limit of permafrost is the weight normalization numerator, and the formula is .

[0102] In a possible implementation, the frozen soil upper limit prediction module 24 may be used to: All trained gradient boosting tree models are combined to obtain the permafrost upper limit fusion prediction model, including: All trained gradient boosting tree models and the weights corresponding to each gradient boosting tree model are multiplied and then summed up to obtain the permafrost upper limit fusion prediction model.

[0103] In a possible implementation, the intrinsic component data determination module 23 may be used to: Calculate the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil, including: Combined with the third preset formula, the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil is calculated. The third preset formula is:

[0104] In the formula, is the Spearman rank correlation coefficient, is the rank difference of each pair of observations in the two variables, is the sample size.

[0105] In a possible implementation, the intrinsic component acquisition module 22 may be used to: The preprocessed permafrost upper limit influencing factor data are subjected to variational mode decomposition of each influencing factor data, including: Combined with the fourth preset formula, the preprocessed frozen soil upper limit influencing factor data are subjected to variational mode decomposition respectively. The fourth preset formula is:

[0106] In the formula, is the Wiener filter of the residual at the current moment, is the number of iterations, ^ is the Fourier transform symbol, is the original signal at frequency The Fourier transform value at It is The modal signal at the frequency The estimated value at is the Lagrange multiplication operator at frequency The value at is the penalty factor, is the frequency, is the center frequency of the component.

[0107] Figure 3 is a schematic diagram of a terminal provided by an embodiment of the present invention. Figure 3 As shown, the terminal 3 of this embodiment includes: a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30. When the processor 30 executes the computer program 32, the steps in the above-mentioned embodiments of the method for predicting the upper limit of permafrost based on modal decomposition are implemented, for example Figure 1 Alternatively, when the processor 30 executes the computer program 32, the functions of the modules in the above-mentioned device embodiments are realized, for example Figure 2 The functions of each module are shown.

[0108] Exemplarily, the computer program 32 may be divided into one or more modules, which are stored in the memory 31 and executed by the processor 30 to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program 32 in the terminal 3. For example, the computer program 32 may be divided into Figure 2 The modules shown.

[0109] The terminal 3 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The terminal 3 may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will appreciate that Figure 3 It is only an example of terminal 3 and does not constitute a limitation on terminal 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal may also include input and output devices, network access devices, buses, etc.

[0110] The processor 30 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0111] The memory 31 may be an internal storage unit of the terminal 3, such as a hard disk or memory of the terminal 3. The memory 31 may also be an external storage device of the terminal 3, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal 3. Further, the memory 31 may also include both an internal storage unit and an external storage device of the terminal 3. The memory 31 is used to store the computer program and other programs and data required by the terminal. The memory 31 may also be used to temporarily store data that has been output or is to be output.

[0112] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0113] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0114] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0115] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0116] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0117] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0118] If the integrated module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned embodiments of the upper limit prediction method of permafrost based on modal decomposition can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable media does not include electrical carrier signals and telecommunication signals.

[0119] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for predicting the upper limit of permafrost based on modal decomposition, characterized in that: include: Obtaining data on factors affecting the upper limit of frozen soil in the area to be predicted; wherein the data on factors affecting the upper limit of frozen soil include temperature, humidity, rainfall and wind speed; Preprocessing the data of factors influencing the upper limit of frozen soil, performing variational mode decomposition on each factor data in the preprocessed data of factors influencing the upper limit of frozen soil, and obtaining the intrinsic components of the data of factors influencing the upper limit of frozen soil; wherein the intrinsic components include periodic terms, trend terms and residual terms; Calculate the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil respectively, and determine the intrinsic component data with high correlation with the upper limit of frozen soil according to the correlation coefficient; The intrinsic component data is input into a pre-trained permafrost upper limit fusion prediction model to obtain a permafrost upper limit prediction result; wherein the permafrost upper limit fusion prediction model is trained based on a pre-constructed data set, the data set includes intrinsic component data with high correlation with historical permafrost upper limits and historical permafrost upper limits, and the permafrost upper limit fusion prediction model includes a gradient boosting sub-model and an adaptive sub-model.

2. The method for predicting the upper limit of permafrost based on modal decomposition according to claim 1 is characterized in that: The gradient boosting sub-model is a gradient boosting tree model, the adaptive sub-model is an adaptive enhancement model, and the training process of the frozen soil upper limit fusion prediction model includes: The error rate of the gradient boosting tree model is calculated according to the intrinsic component data, and the weight of the gradient boosting tree model is calculated according to the error rate of the gradient boosting tree model; the weight of the data set is calculated according to the weight of the adaptive enhancement model and the gradient boosting tree model, the data set is updated according to the weight of the data set, and a new gradient boosting tree model is obtained by training according to the updated data set. When the number of the gradient boosting tree models reaches a preset number, all the trained gradient boosting tree models are combined to obtain the permafrost upper limit fusion prediction model.

3. The method for predicting the upper limit of permafrost based on modal decomposition according to claim 2, characterized in that: The step of calculating the error rate of the gradient boosting tree model according to the intrinsic component data, and calculating the weight of the gradient boosting tree model according to the error rate of the gradient boosting tree model, comprises: Inputting the intrinsic component data with high correlation with the historical upper limit of frozen soil into the gradient boosting tree model to obtain the predicted upper limit of frozen soil; Calculating the error rate of the gradient boosting tree model according to the predicted upper limit of frozen soil and the upper limit of historical frozen soil; Combined with the first preset formula, the weight of the gradient boosting tree model in the frozen soil upper limit fusion prediction model is calculated according to the error rate. The first preset formula is: = In the formula, is the weight of the gradient boosted tree model, is the error rate of the gradient boosted tree model.

4. The method for predicting the upper limit of permafrost based on modal decomposition according to claim 3 is characterized in that: The step of calculating the weight of the data set according to the weights of the adaptive boosting model and the gradient boosting tree model includes: Combined with the second preset formula, the weight of the data set is calculated according to the weight of the gradient boosting tree model, the predicted upper limit of frozen soil, the upper limit of historical frozen soil and the weight of the previous data set. The second preset formula is: In the formula, It is t The weight of the dataset at the iteration, It is i The weights of the gradient boosted tree model, It is a dataset i The historical upper limit of permafrost, It is t Gradient boosted tree model on the dataset i The predicted upper limit of permafrost is the weight normalization numerator, and the formula is .

5. The method for predicting the upper limit of permafrost based on modal decomposition according to claim 4, characterized in that: The method of combining all trained gradient boosting tree models to obtain the frozen soil upper limit fusion prediction model includes: All trained gradient boosting tree models and the weights corresponding to each gradient boosting tree model are multiplied and then summed to obtain the frozen soil upper limit fusion prediction model.

6. The method for predicting the upper limit of permafrost based on modal decomposition according to claim 1, characterized in that: The calculation of the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil includes: Combined with the third preset formula, the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil is calculated. The third preset formula is: In the formula, is the Spearman rank correlation coefficient, is the rank difference of each pair of observations in the two variables, is the sample size.

7. The method for predicting the upper limit of permafrost based on modal decomposition according to claim 1, characterized in that: The method of performing variational mode decomposition on each influencing factor data in the preprocessed permafrost upper limit influencing factor data respectively includes: Combined with the fourth preset formula, the preprocessed frozen soil upper limit influencing factor data are subjected to variational mode decomposition respectively. The fourth preset formula is: In the formula, is the Wiener filter of the residual at the current moment, is the number of iterations, ^ is the Fourier transform symbol, is the original signal at frequency The Fourier transform value at It is The modal signal at the frequency The estimated value at is the Lagrange multiplication operator at frequency The value at is the penalty factor, is the frequency, is the center frequency of the component.

8. A permafrost upper limit prediction device based on modal decomposition, characterized in that: include: A data acquisition module is used to acquire data on factors affecting the upper limit of frozen soil in the area to be predicted; wherein the data on factors affecting the upper limit of frozen soil include temperature, humidity, rainfall and wind speed; An intrinsic component acquisition module is used to preprocess the frozen soil upper limit influencing factor data, and perform variational mode decomposition on each influencing factor data in the preprocessed frozen soil upper limit influencing factor data to obtain the intrinsic components of the frozen soil upper limit influencing factor data; wherein the intrinsic components include periodic terms, trend terms and residual terms; An intrinsic component data determination module is used to calculate the correlation coefficient between the intrinsic component of each influencing factor data and the upper limit of frozen soil, and determine the intrinsic component data with high correlation with the upper limit of frozen soil according to the correlation coefficient; The permafrost upper limit prediction module is used to input the intrinsic component data into a pre-trained permafrost upper limit fusion prediction model to obtain a permafrost upper limit prediction result; wherein the permafrost upper limit fusion prediction model is trained based on a pre-constructed data set, the data set includes intrinsic component data with high correlation with historical permafrost upper limits and historical permafrost upper limits, and the permafrost upper limit fusion prediction model includes a gradient boosting sub-model and an adaptive sub-model.

9. A terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method as claimed in any one of claims 1 to 7 are implemented.

Images