Method and device for predicting deformation of permafrost roadbed and electronic equipment

Abstract

The invention provides a permafrost roadbed deformation prediction method and device and electronic equipment, and belongs to the technical field of permafrost deformation. According to the method, the influence factor data of the frozen soil roadbed can be decomposed, the data structure is simplified, dimension reduction processing is carried out on the decomposed data, and the redundancy-removed principal components are obtained through screening. Then, correlation analysis is carried out on the principal components subjected to redundancy elimination, the principal components with high correlation degree with frozen soil roadbed deformation are extracted, and high-quality input characteristics are provided for a prediction model; according to the method, the variables influencing the deformation of the frozen soil roadbed are screened, key factors with relatively high correlation with the deformation are extracted, data with relatively low correlation are removed, and the quality of a training set is improved. Therefore, on the premise of not influencing the correlation of the original training set, the data volume of the training set is reduced, and the frozen soil roadbed deformation prediction efficiency and accuracy can be improved. Furthermore, the processing advantage of the LSTM on the time sequence characteristics and the efficient calculation characteristic of the XGBoost are fused through the hybrid machine learning model, the calculation cost of frozen soil roadbed deformation prediction can be remarkably reduced while the prediction precision is ensured, and therefore the requirement for efficiency in actual engineering is better met.

Description

Technical Field

[0001] The present invention relates to the technical field of frozen soil deformation, and in particular, to a method, device, and electronic device for predicting the deformation of a permafrost subgrade. Background Art

[0002] Permafrost is a special soil medium, which may exhibit phenomena such as frost heaving, thaw settlement, and the lowering of the permafrost table with climate and seasonal changes, posing a significant threat to the subgrade structure and thus affecting the stability and safety of the road. The stability of the permafrost subgrade is a key issue affecting regional traffic safety and the sustainable development of infrastructure. Therefore, the accurate prediction of the deformation of the frozen soil subgrade is important for the management, prevention, and maintenance of subgrade diseases.

[0003] Traditional methods for predicting frozen soil deformation generally first collect meteorological data (such as temperature, humidity, precipitation, wind speed, etc.), geological data (such as surface temperature, surface heat flux, permafrost table, etc.), and deformation data, and then directly use all the actually collected data for training a permafrost subgrade deformation prediction model to achieve deformation prediction. However, traditional methods for predicting frozen soil deformation are often limited by the complex data of the frozen soil system, the significant non-linear characteristics, and the difficulty in grasping the time-series variation law. For example, the influence degrees of the collected data on the deformation of the permafrost subgrade are different, and a large amount of data with little influence participates in training and prediction, resulting in low accuracy and low prediction efficiency of the permafrost subgrade deformation prediction. Summary of the Invention

[0004] The present invention provides a method, device, and electronic device for predicting the deformation of a permafrost subgrade to solve the problems of low quality of the training set data, low efficiency, and low accuracy of the existing permafrost subgrade deformation prediction.

[0005] In a first aspect, the present invention provides a method for predicting the deformation of a permafrost subgrade, including: Obtaining historical deformation data of the permafrost subgrade and a plurality of variables affecting the deformation of the permafrost subgrade corresponding thereto; Performing variational mode decomposition on any one of the variables to obtain a plurality of modal components of the variable; Based on the kernel principal component analysis method, screening and obtaining a plurality of principal components after removing redundancy for the plurality of modal components of any one of the variables, wherein any one of the principal components is a linear combination of the modal components; Determining the correlation between any one of the principal components and the historical deformation data of the permafrost subgrade; Taking the principal components with a correlation greater than a threshold value among the principal components as a key factor data set; Using the key factor data set and the historical deformation data of the permafrost subgrade as a training set to perform neural network training to obtain a permafrost subgrade deformation prediction model; Based on the frozen soil subgrade deformation prediction model, predict the deformation of the frozen soil subgrade.

[0006] In a possible implementation manner, the types of the modal components include periodic terms, trend terms, and residual terms.

[0007] In a possible implementation manner, after obtaining multiple modal components of the variable, it further includes: For each modal component, based on the maximum-minimum normalization, perform a linear transformation on the modal component to obtain a normalized component; Use the normalized component as the modal component.

[0008] In a possible implementation manner, the frozen soil subgrade deformation prediction model includes a fusion algorithm; the fusion algorithm is formed by combining multiple algorithms.

[0009] In a possible implementation manner, the frozen soil subgrade deformation prediction model includes a long short-term memory network and a gradient boosting decision tree.

[0010] In a possible implementation manner, based on the frozen soil subgrade deformation prediction model, predicting the deformation of the frozen soil subgrade includes: Obtain the key factor data affecting the deformation of the frozen soil subgrade; The long short-term memory network extracts the time-dependent features in the key factor data and outputs them to the gradient boosting decision tree; The gradient boosting decision tree uses the static distribution information of the time-dependent features to perform regression or classification tasks and generates a prediction result for the deformation of the frozen soil subgrade.

[0011] In a possible implementation manner, for the multiple modal components of any one of the variables, screening to obtain multiple principal components after removing redundancy based on the kernel principal component analysis method includes: Based on the multiple modal components of any one of the variables, construct an input matrix; According to the input matrix, based on the kernel principal component analysis method, obtain multiple principal components, where any one of the principal components is a linear combination of each modal component; For any one of the principal components, determine the contribution rate of the principal component; Arrange the contribution rates of each principal component in descending order, and select the principal components with contribution rates greater than the threshold as the principal components after removing redundancy.

[0012] In a possible implementation manner, determining the correlation between any one of the principal components and the historical deformation data of the frozen soil subgrade includes: Determine the correlation based on the following formula:

[0013] Among them, represents the correlation; represents the value of the th principal component; represents the average value of the th principal component; represents the th deformation value; represents the average deformation value.

[0014] In a second aspect, the present invention provides a device for predicting the deformation of a frozen soil subgrade, including: An acquisition module, configured to acquire historical deformation data of the frozen soil subgrade and a plurality of variables affecting the deformation of the frozen soil subgrade corresponding thereto; A decomposition module, configured to perform variational mode decomposition on any one of the variables to obtain a plurality of modal components of the variable; A screening module, configured to screen, based on the kernel principal component analysis method, a plurality of principal components after removing redundancy from the plurality of modal components of any one of the variables, wherein any one of the principal components is a linear combination of the modal components; A correlation determination module, configured to determine the correlation between any one of the principal components and the historical deformation data of the frozen soil subgrade; A judgment module, configured to use the principal components with a correlation greater than a threshold among the principal components as a key factor data set; A training module, configured to use the key factor data set and the historical deformation data of the frozen soil subgrade as a training set to perform neural network training to obtain a frozen soil subgrade deformation prediction model; A prediction module, configured to predict the deformation of the frozen soil subgrade based on the frozen soil subgrade deformation prediction model.

[0015] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method described in the first aspect or any one of the possible implementation manners of the first aspect are implemented.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the method described in the first aspect or any one of the possible implementation manners of the first aspect are implemented.

[0017] The present invention provides a method, device and electronic equipment for predicting the deformation of permafrost subgrade. By decomposing the influencing factor data of the permafrost subgrade and simplifying the data structure, the present invention then performs dimensionality reduction processing on the decomposed data, and screens to obtain the principal components after removing redundancy. Then, correlation analysis is performed on the principal components after removing redundancy, and the principal components with a relatively high degree of correlation with the deformation of the permafrost subgrade are extracted, providing high-quality input features for the prediction model. The present invention screens the variables affecting the deformation of the permafrost subgrade, extracts the key factors with a strong correlation with the deformation, removes the data with a low correlation, and improves the quality of the training set. Thus, on the premise of not affecting the correlation of the original training set, the key factor data set is screened and determined, the amount of training set data is reduced, and the prediction efficiency and accuracy of the permafrost subgrade deformation can be improved. Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is an application scenario diagram of the method for predicting the deformation of permafrost subgrade provided by the embodiment of the present invention; Figure 2 It is a flowchart of the implementation of the method for predicting the deformation of permafrost subgrade provided by the embodiment of the present invention; Figure 3 It is a flowchart of the implementation of the data processing method provided by the embodiment of the present invention; Figure 4 It is a schematic diagram of variational mode decomposition provided by the embodiment of the present invention; Figure 5 It is a schematic diagram of kernel principal component analysis method provided by the embodiment of the present invention; Figure 6 It is a heat map of the correlation between the principal components and the deformation provided by the embodiment of the present invention; Figure 7 It is a schematic diagram of the structure of the permafrost subgrade deformation prediction model provided by the embodiment of the present invention; Figure 8 It is a schematic diagram of the prediction result of the LSTM model based on hyperparameter adjustment provided by the embodiment of the present invention; Figure 9 It is a schematic diagram of the prediction result of the hybrid model based on hyperparameter adjustment provided by the embodiment of the present invention; Figure 10 It is a heat map of the correlation between the principal components and the deformation of the second point provided by the embodiment of the present invention; Figure 11It is a schematic diagram of the prediction result of the LSTM model based on hyperparameter adjustment provided by the embodiment of the present invention; Figure 12 It is a schematic diagram of the prediction result of the hybrid model based on hyperparameter adjustment provided by the embodiment of the present invention; Figure 13 It is a schematic diagram of the structure of the frozen soil subgrade deformation prediction device provided by the embodiment of the present invention; Figure 14 It is a schematic diagram of the electronic device provided by the embodiment of the present invention. Detailed implementation manners

[0020] In the following description, specific details such as specific system structures and technologies are proposed for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present invention. However, those skilled in the art should clearly understand that the present invention can also be implemented in other embodiments without these specific details. In other cases, the detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from hindering the description of the present invention.

[0021] To make the purpose, technical solutions, and advantages of the present invention clearer, the following will be described through specific embodiments in conjunction with the accompanying drawings.

[0022] Figure 1 It is an application scenario diagram of the permafrost subgrade deformation prediction method provided by the embodiment of the present invention. As Figure 1 shown, the lower layer of the frozen soil subgrade includes a frozen soil layer. Frozen soil subgrades are generally found in plateau areas. With the continuous development of computer science and artificial intelligence technologies, the prediction method of frozen soil subgrades based on advanced algorithms has gradually become a research hotspot.

[0023] Currently, traditional frozen soil deformation prediction methods are often limited due to the complex data of the frozen soil system, significant non-linear characteristics, and difficult-to-grasp time series change laws. Moreover, due to the greater complexity of the plateau environment, the acquired data may have missing or incorrect data mixed in. Ordinary data processing methods are difficult to extract high-quality data.

[0024] There are many influencing factors for frozen soil subgrade deformation, such as meteorological factors, geological factors, etc. In the existing prediction methods, generally all the actually collected data are directly used to train the frozen soil subgrade deformation prediction model, and the accuracy and prediction efficiency of this method are relatively low. On the one hand, the influencing degrees of the existing collected factors on the frozen soil subgrade deformation are different. On the other hand, there is also a correlation between the factor data, which leads to data duplication.

[0025] The present invention provides a method for predicting the deformation of permafrost subgrade. By screening and processing the data of influencing factors of the permafrost subgrade, high-quality input features are provided for the prediction model to solve the problems of low efficiency and low accuracy in the existing prediction of permafrost subgrade deformation due to the low quality of the training set data.

[0026] Figure 2 FIG. is a flowchart for implementing the method for predicting the deformation of permafrost subgrade provided by an embodiment of the present invention. Referring to Figure 2 , it is described in detail as follows: In step 201, historical deformation data of the permafrost subgrade and a plurality of variables affecting the deformation of the permafrost subgrade are obtained; Figure 3 FIG. is a flowchart for implementing the data processing method provided by an embodiment of the present invention; referring to Figure 3 , it mainly shows the implementation process of steps 201-205. Exemplarily, the historical deformation data of the permafrost subgrade may be time series data. The plurality of variables affecting the deformation of the permafrost subgrade may also be time series data. Further, there is a temporal correspondence between the historical deformation data of the permafrost subgrade and each variable affecting the deformation of the permafrost subgrade.

[0027] The variables affecting the deformation of the permafrost subgrade may also be referred to as influencing factors. In some embodiments, the variables affecting the deformation of the permafrost subgrade include the air temperature, humidity, precipitation, wind speed at 2 m, surface temperature, surface heat flux, or permafrost table at 2 m. It should be noted that the variables here may be variables composed of a set of time series data. For example, for a certain cross-section of the permafrost subgrade, the humidity data at each moment within a period of time at this position can form a humidity variable.

[0028] Exemplarily, the meteorological data of this area can be obtained through the meteorological bureau. For example, the temperature at 2 m, the air humidity at 2 m, precipitation, wind speed at 2 m, and surface temperature, etc. Another example is that sensors with different depths can be arranged in the subgrade to collect the deformation data of the permafrost subgrade.

[0029] Exemplarily, the data obtained above can be sorted out and imported into the original database. It should be noted that the original data obtained above often cannot be directly used and needs to be preprocessed. The preprocessing usually includes missing value processing, outlier detection and processing, data cleaning, and data normalization processing, etc.

[0030] In step 202, for any variable, variational mode decomposition is performed to obtain a plurality of modal components of the variable; Figure 4 FIG. is a schematic diagram of variational mode decomposition provided by an embodiment of the present invention; referring to Figure 4, Exemplarily, modal decomposition is performed for each variable to obtain multiple modal components of the variable. For example, for 5 variables, each variable is decomposed into 3 modal components, resulting in a total of 15 modal components.

[0031] VMD (Variational Mode Decomposition) is an adaptive quasi-orthogonal signal decomposition method proposed based on classical modal decomposition methods. The VMD algorithm decomposes the original signal into several intrinsic mode functions with practical physical meanings by constructing a variational problem and iterative solution. Most of the decomposed sub-signals pulsate around the center, and VMD can be regarded as a constrained variational problem, that is

[0032] where: u k is the modal component; ω k is the center frequency of each component; δ(t) is the Dirac distribution with respect to time t; e -jωkt is the estimated center frequency of each analytic signal; k is the modal index of decomposition; ∂ t is the partial derivative; f is the original signal.

[0033] To solve the above constrained variational model, a quadratic penalty term and a Lagrange multiplier are introduced, and the above equations (1) and (2) are rewritten as:

[0034] L(·) is the augmented Lagrangian function; ‖•‖ 2 2 represents the L2 norm, that is, the Euclidean norm of the vector; 〈•〉 is the symbol for calculating the inner product, and λ is the Lagrange multiplier. Equation (3) can be solved using the alternating direction multiplier algorithm for the saddle point in the equation, and then u k and ω k are updated according to equations (4) and (5):

[0035]

[0036] where: α is the bandwidth balance parameter; i (ω), k (ω), k (ω) and k (ω) are respectively i (t), μ k(t), f(t), and the function obtained by Fourier transforming λ(t) can, through Equations (3) to (5), decompose the original signal f into K intrinsic mode functions (IMFs), each IMF having different scale characteristics and reflecting the respective essential information in the original signal.

[0037] The process of the VMD model is as follows: Step ①: Initialization , , and n.

[0038] Step ②: Update and .

[0039] Step ③: Use to update , where represents the time-delay operator.

[0040] Step ④: Set the judgment accuracy e > 0. If the judgment is not satisfied, return to Step ②; if the judgment is satisfied, stop the iteration.

[0041] Exemplarily, variational mode decomposition can be performed one by one for any one of the variables such as air temperature, humidity, precipitation, 2-meter wind speed, surface temperature, surface heat flux, or permafrost table at 2 meters.

[0042] In the subsequent search and solution process, the optimal central frequency and finite bandwidth of each mode can be adaptively matched, and the effective separation of the intrinsic mode components (IMFs), the frequency-domain division of the signal, and thus the effective decomposition components of the given signal can be achieved, and finally the optimal solution of the variational problem can be obtained.

[0043] To effectively capture the different frequency components of the influencing factors, several key parameters of VMD need to be set, including the number of modes K, the bandwidth constraint parameter α, and the initialization method. The setting of these parameters not only affects the effect of VMD decomposition but is also indirectly related to permafrost deformation. Exemplarily, the number of modes can be 3. Exemplarily, in dealing with the problem of permafrost deformation, each of the five groups of influencing factors is decomposed into three groups of data, IMF1, 2, and 3, that is, the original five groups of data are decomposed into fifteen groups of influencing factors after decomposition. The following is an example.

[0044] In one possible implementation, the types of modal components include periodic terms, trend terms, and residual terms.

[0045] Exemplarily, the periodic term represents the component that varies periodically in the variable; the trend term represents the long-term trend change in the variable; and the residual term represents the remaining part after removing the periodic term and the trend term from the variable.

[0046] In the embodiment of the present invention, the variational mode decomposition (VMD) technology is used to decompose complex data into a periodic term, a trend term, and a residual term, thereby effectively simplifying the data structure.

[0047] Data standardization is a preprocessing technology aimed at preventing certain features from overly influencing other features. The goal of data standardization is to ensure that features sharing the same scale contribute equally to the model. The following provides a standardization method to process each component obtained in step 202.

[0048] In a possible implementation manner, after obtaining multiple modal components of the variable, it further includes: for each modal component, based on the maximum-minimum normalization, performing a linear transformation on the modal component to obtain a normalized component; and using the normalized component as the modal component.

[0049] Exemplarily, the following formula is used for normalization processing:

[0050] where x' represents the normalized value, x represents the original feature value before normalization, x min represents the minimum value of the feature, and x max represents the maximum value of the feature. After normalization processing, all feature ranges are standardized to fall within the range of [0 - 1]. Replace the component before normalization with the normalized component to update the modal component.

[0051] In step 203, for multiple modal components of any variable, based on the kernel principal component analysis method, multiple principal components after removing redundancy are screened, where any principal component is a linear combination of each modal component; Figure 5 is a schematic diagram of the kernel principal component analysis method provided by the embodiment of the present invention; referring to Figure 5 , exemplarily, taking all modal components of any variable as a whole and performing kernel principal component analysis, multiple principal components can be obtained. Each principal component is a linear combination of all modal components.

[0052] It should be noted that among the modal components of each variable obtained in step 202, the regular information represented by each component may be repeated. Using the repeated information as the basis for predicting the deformation of the frozen soil subgrade greatly reduces the efficiency. Through the kernel principal component analysis (KPCA) in the embodiment of the present invention, the modal components of each variable can be classified and the problem of variable redundancy can be solved.

[0053] It should also be noted that each obtained principal component is a linear combination of the modal components of a certain variable. Therefore, the purpose here is not to remove a certain variable or a certain component. Instead, it is to remove the redundant parts in each component and retain the information after redundancy removal. The following embodiments illustrate the specific implementation steps of KPCA.

[0054] Kernel principal component analysis (KPCA) is used to classify data and solve the problem of variable redundancy. During the prediction process, the data after variational mode decomposition is mainly mapped from high dimension to low dimension.

[0055] In a possible implementation manner, for multiple modal components of any variable, based on the kernel principal component analysis method, screening to obtain multiple principal components after redundancy removal includes: Step 2031: Based on multiple modal components of any variable, construct an input matrix; It should be noted that the modal components of any variable can be a set of time series data, and the time dimensions of each component can be the same. For example, each component has 1 data point every 5 minutes.

[0056] Exemplarily, an input matrix in the following form can be constructed:

[0057] where x n represents the nth component. Each component has m data points, and an m×n-dimensional matrix X can be formed as the input space.

[0058] Step 2032: According to the input matrix, based on the kernel principal component analysis method, obtain multiple principal components, where any principal component is a linear combination of the modal components; Exemplarily, X is mapped to a higher-dimensional feature space F. After centralized processing, X′ is obtained, k is determined, and the eigenvalue λ is solved. X′ is the projection of the eigenvector (p) on F:

[0059] where represents the central matrix, represents an n-dimensional column vector: = 1, 2, 3,..., n]T.

[0060] Step 2033: For any principal component, determine the contribution rate of this principal component; Exemplarily, the contribution rate of the principal component is determined based on the following formula:

[0061] Among them, L represents the contribution rate.

[0062] Step 2034: Arrange the contribution rates of the principal components in descending order, and select the principal components with contribution rates greater than the threshold as the principal components after removing redundancy.

[0063] Alternatively, the principal components with cumulative contribution rates greater than the set value can also be used as the principal components after removing redundancy. Among them, the cumulative contribution rate of the principal component is determined based on the following formula:

[0064] Among them, K represents the cumulative contribution rate.

[0065] In the embodiment of the present invention, kernel principal component analysis (KPCA) is used to perform dimensionality reduction processing on the decomposed data, which can further reduce the dimension and improve the calculation efficiency. The following will further explain and combine with correlation analysis to extract the influencing factors most relevant to the subgrade deformation, and provide high-quality input features for the prediction model.

[0066] In step 204, for any principal component, determine the correlation between the principal component and the historical deformation data of the frozen soil subgrade; Exemplarily, in step 204, the principal components screened in step 203 are targeted. The screened principal components are the remaining principal components after removing the redundant principal components among all the principal components.

[0067] It should be noted that among the principal components after removing redundancy in the previous step, the degree of correlation between each principal component and the deformation is different. This will lead to the fact that if the principal components with low correlation are used for deformation prediction, the prediction accuracy will not be improved, but the efficiency will decrease instead. Therefore, in the embodiment of the present invention, by screening out the principal components with high correlation, the efficiency and accuracy of the final frozen soil subgrade deformation prediction can be improved.

[0068] Correlation is the linear relationship between two or more variables. Based on correlation, one variable can be predicted according to another variable. The basic principle of using correlation in feature selection is based on the idea that meaningful variables show robust correlation with the result: there must be a correlation between the variable and the deformation. The higher the correlation, the greater the influence of the variable on the deformation prediction result.

[0069] In a possible implementation manner, determining the correlation between any principal component and the historical deformation data of the frozen soil subgrade includes: Determine the correlation based on the following formula:

[0070] Among them, represents the correlation; represents the value of the th principal component; represents the average value of the th principal component; represents the th deformation value; represents the average deformation value.

[0071] In some embodiments, a heat map can be used to visualize the correlation between features in a dataset.

[0072] Figure 6 is the heat map of the correlation between the principal component and the deformation provided by the embodiment of the present invention; refer to Figure 6 , the horizontal axis and the vertical axis are 11 elements, including 10 principal components and 1 deformation. The value at the intersection of the row and the column in the figure represents the correlation between the two elements. Among them, the last row and the last column are the correlations between the principal components and the deformation. The bottom row of the vertical axis and the rightmost column of the horizontal axis represent the frozen soil deformation. PC1(AT), PC2(AT), PC1(ST), PC2(ST), PC1(P), PC2(P), PC1(SH), PC2(SH), PC1 and PC2 represent the respective principal components.

[0073] A positive correlation relationship is observed between some elements, such as the data of PC2(ST) and the frozen soil deformation, and PC1(AT) and PC1(ST). On the contrary, a negative relationship is also obvious, such as the relationship between PC1(P) and the frozen soil deformation. It can be seen from the figure that the correlation between PC1(P) and the frozen soil deformation is very strong, which is 0.75.

[0074] In step 205, the principal components with a correlation greater than the threshold in each principal component are used as the key factor dataset; Exemplarily, the principal components with a correlation not greater than the threshold are removed, and the principal components with a correlation greater than the threshold are retained to form the key factor dataset.

[0075] It should be noted that the greater the correlation, the more relevant the principal component is to the deformation.

[0076] By performing variational mode decomposition (VMD) and kernel principal component analysis (KPCA) on the original data and combining correlation analysis to extract key factors with strong correlations, the embodiment of the present invention significantly improves the effectiveness of features.

[0077] In step 206, the key factor dataset and the historical deformation data of the frozen soil subgrade are used as the training set for neural network training to obtain the frozen soil subgrade deformation prediction model; It should be noted that the key factor dataset obtained in step 205 is data that has been de-duplicated and screened for relevance, and can provide high-quality input features for the prediction model. In terms of data form, it is no longer the form of the original variables that affect the deformation of frozen soil subgrade in step 201. The prediction model of frozen soil subgrade deformation will be described below.

[0078] In a possible implementation, the prediction model of frozen soil subgrade deformation includes a fusion algorithm; the fusion algorithm is formed by combining multiple algorithms.

[0079] In the embodiment of the present invention, a fusion algorithm is formed by combining multiple algorithms, which can give play to the advantages of each algorithm and improve the accuracy, reliability and adaptability of the prediction of frozen soil subgrade deformation.

[0080] In a possible implementation, the prediction model of frozen soil subgrade deformation includes a long short-term memory network and a gradient boosting decision tree.

[0081] The long short-term memory network (LSTM) is a special form of time recurrent neural network, aiming to solve the inherent long-term dependence problem of traditional recurrent neural networks (RNNs). Different from the standard RNN, the LSTM network replaces the hidden layer neurons with memory units. These memory units are composed of an input gate, a forget gate and an output gate, allowing the network to selectively retain or discard information at each time step. This design effectively solves the difficulty of long-term preservation of relevant data. The LSTM recurrent network has been widely recognized for its ability to capture temporal correlations and has been proven to be particularly effective in different fields such as language translation and speech recognition. In predicting the deformation of frozen soil, since the deformation of frozen soil changes over time, it has temporality. It has been verified that LSTM can well analyze its prediction changes and make good predictions.

[0082] However, in existing research, although the long short-term memory network (LSTM) has shown excellent results in processing temporal data prediction tasks, the training process of LSTM is usually accompanied by high computational costs and time consumption, especially in the hyperparameter tuning stage. This high computational cost limits its efficient application in actual engineering scenarios.

[0083] XGBoost is an efficient gradient boosting tree model, known for its excellent performance and flexibility. It achieves fast training through parallel computing and hardware optimization, supports multiple objective functions and data formats, and has a powerful regularization mechanism that can effectively prevent overfitting. Its ability to automatically handle missing values, support weighted data distributions and incremental learning makes it perform well in unbalanced data and dynamic data scenarios. In addition, XGBoost provides feature importance scores, which is convenient for feature engineering and can be seamlessly integrated with mainstream tools, and is widely used in industrial practice and machine learning competitions.

[0084] Although tree-based models such as XGBoost perform well in many machine learning tasks, they have significant limitations when dealing with time series data prediction tasks. Tree models mainly rely on static partitioning and modeling of features, and their core algorithms are not optimized for the time dependence and sequential characteristics in the data. Therefore, when dealing with data with significant temporal relationships, tree models may have difficulty effectively capturing dynamic trends, periodic patterns, or long-term dependence characteristics in time series. In addition, due to the lack of inherent modeling ability for temporal features, using tree models alone may require a large amount of additional manual feature engineering to introduce time information artificially, which not only increases complexity but may also lead to a decline in model performance. These disadvantages make tree models unable to fully meet the actual needs when dealing with complex time series prediction tasks.

[0085] In order to improve the computational efficiency of the frozen soil subgrade deformation prediction model in practical applications, the embodiments of the present invention combine LSTM with the gradient boosting decision tree (XGBoost) model. As a tree-based model, XGBoost not only has strong generalization ability and robustness but is also known for its fast training speed and low computational overhead. By integrating the modeling advantages of LSTM for temporal features and the efficient computational characteristics of XGBoost, it is possible to significantly reduce the computational cost of the model while ensuring prediction accuracy, thus better meeting the efficiency requirements in actual engineering. This combination method provides a new idea for constructing an efficient and accurate time series prediction model.

[0086] In the construction of the prediction model in the embodiments of the present invention, the LSTM-XGBoost fusion algorithm is adopted to meet the dual requirements of time dependence modeling and computational efficiency in frozen soil deformation prediction. The long short-term memory network (LSTM) has the advantage of capturing long-term and short-term dependence relationships in time series data and can fully explore the dynamic change characteristics of the frozen soil system; the gradient boosting decision tree (XGBoost) is characterized by its fast and efficient computational ability and strong non-linear modeling performance, and has outstanding adaptability and application value in engineering practice. By combining the two, the LSTM-XGBoost fusion model can not only significantly improve the prediction accuracy but also effectively reduce the computational cost, providing an efficient and reliable solution for frozen soil subgrade deformation prediction.

[0087] Figure 7 is a schematic structural diagram of the frozen soil subgrade deformation prediction model provided by the embodiments of the present invention; referring to Figure 7 , the frozen soil subgrade deformation prediction model includes an input layer, a hidden layer, and an output layer.

[0088] Figure 8 is a schematic diagram of the prediction result of the LSTM model based on hyperparameter tuning provided by the embodiments of the present invention; referring to Figure 8, the horizontal axis is time and the vertical axis is the deformation of the frozen soil subgrade. The dot line is the actually measured deformation value, and the cross line is the predicted deformation value. Among them, the evaluation indexes of the model: RMSE (root mean square error) is 0.06532847444842008; R² (coefficient of determination) is 0.9779509204436425.

[0089] Figure 9 is a schematic diagram of the prediction result of the hybrid model based on hyperparameter adjustment provided by an embodiment of the present invention; referring to Figure 9 , the hybrid model refers to the LSTM-XGBoost model. Among them, the evaluation indexes of the hybrid model: RMSE: 0.13381388867853825; R²: 0.9074901845828327.

[0090] The LSTM-XGBoost model is a hybrid model design method that combines the advantages of deep learning and traditional machine learning, aiming to solve the limitations of a single algorithm in time series prediction tasks.

[0091] Specifically, this method extracts and models time series data through a long short-term memory network (LSTM), fully capturing the time dependence and complex dynamic patterns in the data; then takes the extracted high-dimensional time series features as input and further hands them over to XGBoost for efficient regression or classification tasks. XGBoost is responsible for making full use of the static distribution information of the features in this process, quickly generating accurate prediction results, and at the same time avoiding the overfitting or computational efficiency problems that may occur when the LSTM model directly regresses in the output layer.

[0092] This fusion method not only retains the deep learning advantage of LSTM in modeling time series features, but also utilizes the fast and efficient computational characteristics of XGBoost, enabling the model to achieve a good balance between prediction accuracy and computational efficiency, and providing a flexible and practical solution for time series prediction tasks.

[0093] Among them, the specific methods for implementing the fusion of machine algorithms include: simple average method, weighted average model output, weighted average model parameters, voting method, bagging, stacking method (Stacking), boosting method (Boosting), and GBDT, etc. After comparing various methods, the stacking method is finally selected. Stacking is a more complex model aggregation method that takes the prediction results of multiple different models as input and trains a new model to generate the final prediction. This method can effectively utilize the prediction capabilities of different models.

[0094] Stacking is an ensemble learning technique that combines multiple different base models and uses a meta-model (or secondary learner) to integrate the prediction results of each base model to obtain better prediction performance.

[0095] After actual calculation and verification, the LSTM-XGBoost model has shown significant advantages in terms of computational efficiency. Specifically, the computational time of this algorithm is only 34.1 seconds. Compared with the 22 minutes (1320 seconds) of the ordinary LSTM algorithm, the time efficiency has increased by nearly 3770.96%, fully demonstrating the superiority of the hybrid model in reducing computational costs.

[0096] In addition, the LSTM-XGBoost model also shows stable performance in prediction. Its R² index can stably reach above 0.9, proving its high applicability and accuracy for complex time series data prediction tasks. Such results indicate that the LSTM-XGBoost fusion model can not only meet the requirements of prediction accuracy in actual engineering scenarios, but also significantly reduce time and resource costs, providing a practical solution for building an efficient and reliable time series prediction system.

[0097] The following explains the hyperparameter tuning of the embodiments of the present invention. Hyperparameter adjustment is a process of optimizing the performance of the application model for predicting frozen soil deformation and fine-tuning the model parameters. This is crucial for achieving the goal. The adjustments include: hyperparameter tuning of the LSTM model and hyperparameter tuning of the XGBoost model.

[0098] The parameters for hyperparameter tuning of the LSTM model include: units (number of neurons), activation (activation function), and optimizer (optimizer). Exemplarily, the best parameters can be selected by the minimum loss value of the validation set.

[0099] The hyperparameter tuning of the XGBoost model includes: using the method of grid search (GridSearchCV) to determine n_estimators (number of trees), max_depth (depth of the tree), and learning_rate (learning rate). Exemplarily, cross-validation can also be used to determine the best parameters.

[0100] The embodiment of the present invention is based on a hybrid machine learning model combining LSTM and XGBoost for predicting frozen soil deformation. Given the time series characteristics of the frozen soil deformation problem, experimental results show that the LSTM model performs best in terms of prediction accuracy. However, the training of the LSTM model is usually accompanied by high computational costs and time consumption, especially in the hyperparameter tuning stage, which becomes a bottleneck for its popularization and application in actual engineering scenarios. To solve the above problems, the embodiment of the present invention combines LSTM with the XGBoost model with higher computational efficiency to construct an LSTM-XGBoost hybrid model. While maintaining high prediction accuracy, this model significantly reduces the computational cost.

[0101] In step 207, based on the frozen soil subgrade deformation prediction model, the frozen soil subgrade deformation is predicted.

[0102] In a possible implementation manner, predicting the frozen soil subgrade deformation based on the frozen soil subgrade deformation prediction model includes: Obtain the key factor data affecting the frozen soil subgrade deformation; The long short-term memory network extracts the time-dependent features in the key factor data and outputs them to the gradient boosting decision tree; The gradient boosting decision tree uses the static distribution information of the time-dependent features to perform regression or classification tasks and generates the prediction result of the frozen soil subgrade deformation.

[0103] In some embodiments, obtaining the key factor data affecting the frozen soil subgrade deformation includes: Step A: Obtain multiple variables affecting the frozen soil subgrade deformation.

[0104] It should be noted that the variables obtained here are the same as those obtained in step 201.

[0105] Step B: For any variable, perform variational mode decomposition to obtain multiple modal components of the variable.

[0106] It should be noted that the same method as in step 202 is used for modal decomposition here. For example, the number of decomposition terms is the same, and the central frequencies of each term are the same.

[0107] Step C: For the modal components of each variable, obtain the principal components, where any principal component is a linear combination of the modal components.

[0108] It should be noted that the linear relationships between the principal components and the modal components are also the same as those in step 203. That is, the relationships between the principal components and the modal components are the linear relationships determined in step 203. It should also be noted that the principal components to be obtained in step C are the principal components after screening and removing redundancy in step 203, and at the same time, they also need to be the principal components with a correlation greater than the threshold in step 205. That is, the principal components to be obtained here are the principal components in the training set.

[0109] Step D: Form the key factor data affecting the deformation of the frozen soil subgrade from the principal components obtained in step C, and use it as the input of the frozen soil subgrade deformation prediction model.

[0110] In the embodiment of the present invention, by performing variational mode decomposition (VMD) and kernel principal component analysis (KPCA) on the original data, and combining correlation analysis to extract key factors with strong correlations, the effectiveness of the input features is significantly improved.

[0111] In the embodiment of the present invention, the influencing factor data of the frozen soil subgrade is decomposed to simplify the data structure, and then the dimensionality reduction process is performed on the decomposed data to screen and obtain the principal components after removing redundancy. Then, correlation analysis is performed on the principal components after removing redundancy to extract the principal components with a relatively high degree of correlation with the deformation of the frozen soil subgrade, providing high-quality input features for the prediction model. The present invention screens the variables affecting the deformation of the frozen soil subgrade, extracts the key factors with strong correlations with the deformation, removes the data with low correlations, and improves the quality of the training set. Thus, without affecting the correlation of the original training set, the amount of training set data is reduced, and the prediction efficiency and accuracy of the frozen soil subgrade deformation can be improved.

[0112] The embodiment of the present invention provides an efficient and reliable solution for frozen soil deformation prediction, and at the same time provides reference and inspiration for the research of other complex nonlinear time series problems.

[0113] In the embodiment of the present invention, to verify the generalization ability of the frozen soil deformation prediction model, the frozen soil deformation prediction model developed based on the first point of the frozen soil subgrade section is applied to the second point on the section for cross-validation.

[0114] In the preliminary verification, the frozen soil deformation prediction model shows good performance for the first point on the section. To further confirm the applicability of the frozen soil deformation prediction model, the second point on the section is selected, and the generalization ability of its algorithm is tested using the same modeling method and verification process. The experimental results show that the frozen soil deformation prediction model still achieves relatively excellent performance at the new section point, verifying its universality and generalization ability at different section points. This result provides a solid foundation for the further promotion and application of this frozen soil deformation prediction model.

[0115] Figure 10It is the heat map of the correlation between the main components and the deformation of the second point provided by the embodiment of the present invention; Figure 11 It is the schematic diagram of the prediction result of the LSTM model based on hyperparameter adjustment of the second point provided by the embodiment of the present invention; Refer to Figure 11 , the evaluation indexes of the model: RMSE (root mean square error): 0.04366695181593481; R² (coefficient of determination): 0.9496204242759894.

[0116] Figure 12 It is the schematic diagram of the prediction result of the hybrid model based on hyperparameter adjustment of the second point provided by the embodiment of the present invention; Refer to Figure 12 , the evaluation indexes of the hybrid model: RMSE: 0.07736477451229669; the coefficient of determination R² of the hybrid model: 0.8418623502405721.

[0117] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0118] The following is the device embodiment of the present invention. For the details not described in detail, reference can be made to the corresponding method embodiments above.

[0119] Figure 13 It is the structural schematic diagram of the frozen soil subgrade deformation prediction device provided by the embodiment of the present invention. For the convenience of description, only the parts related to the embodiment of the present invention are shown and are described in detail as follows: Refer to Figure 13 , the frozen soil subgrade deformation prediction device 13 includes: An acquisition module 131, configured to acquire the historical deformation data of the frozen soil subgrade and a plurality of variables corresponding to the deformation of the frozen soil subgrade; A decomposition module 132, configured to perform variational mode decomposition on any variable to obtain a plurality of modal components of the variable; A screening module 133, configured to screen and obtain a plurality of principal components after removing redundancy based on kernel principal component analysis for the plurality of modal components of any variable, wherein any principal component is a linear combination of each modal component; A correlation determination module 134, configured to determine the correlation between any principal component and the historical deformation data of the frozen soil subgrade; A judgment module 135, configured to use the principal components with a correlation greater than a threshold among the principal components as the key factor data set; A training module 136, configured to use the key factor data set and the historical deformation data of the frozen soil subgrade as a training set to perform neural network training to obtain a frozen soil subgrade deformation prediction model; A prediction module 137, configured to predict the deformation of the frozen soil subgrade based on a frozen soil subgrade deformation prediction model.

[0120] In the embodiment of the present invention, the influence factor data of the frozen soil subgrade is decomposed to simplify the data structure, and then the dimension reduction processing is performed on the decomposed data, and the main components after redundancy removal are screened out. Then, the correlation analysis is performed on the main components after redundancy removal, and the main components with a relatively high correlation with the deformation of the frozen soil subgrade are extracted, providing high-quality input features for the prediction model. The present invention screens the variables affecting the deformation of the frozen soil subgrade, extracts the key factors with a strong correlation with the deformation, removes the data with a low correlation, and improves the quality of the training set. Thus, without affecting the correlation of the original training set, the data volume of the training set is reduced, and the prediction efficiency and accuracy of the frozen soil subgrade deformation can be improved.

[0121] Figure 14 is a schematic diagram of an electronic device provided by an embodiment of the present invention. As Figure 14 shown, the electronic device 14 of this embodiment includes: a processor 140, a memory 141, and a computer program 142 stored in the memory 141 and executable on the processor 140. When the processor 140 executes the computer program 142, the steps in the above-mentioned embodiments of various permafrost subgrade deformation prediction methods are implemented, such as Figure 2 the steps 201 to 207 shown. Alternatively, when the processor 140 executes the computer program 142, the functions of each module / unit in the above-mentioned device embodiments are implemented, such as Figure 13 the functions of the modules 131 to 137 shown.

[0122] Exemplarily, the computer program 142 can be divided into one or more modules / units, and the one or more modules / units are stored in the memory 141 and executed by the processor 140 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program 142 in the electronic device 14. For example, the computer program 142 can be divided into Figure 13 the modules 131 to 137 shown.

[0123] The electronic device 14 can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The electronic device 14 may include, but is not limited to, a processor 140 and a memory 141. Those skilled in the art can understand, Figure 14This is only an example of the electronic device 14, which does not constitute a limitation on the electronic device 14. It may include more or fewer components than those shown in the figure, or combine certain components, or have different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0124] The so-called processor 140 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0125] The memory 141 may be an internal storage unit of the electronic device 14, such as the hard disk or memory of the electronic device 14. The memory 141 may also be an external storage device of the electronic device 14, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 14. Further, the memory 141 may also include both the internal storage unit and the external storage device of the electronic device 14. The memory 141 is used to store the computer program and other programs and data required by the electronic device. The memory 141 may also be used to temporarily store data that has been output or is to be output.

[0126] Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules according to needs, 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. Each functional unit and module in the embodiment can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of this application. The specific working process of the units and modules in the above system can refer to the corresponding process in the foregoing method embodiment and will not be elaborated herein.

[0127] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0128] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0129] In the embodiments provided by the present invention, it should be understood that the disclosed device / electronic device and method can be implemented in other ways. For example, the device / electronic device embodiments described above are only illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there can be other division methods. For example, 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 displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical or other forms.

[0130] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0131] In addition, in each embodiment of the present invention, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0132] If the integrated module / unit 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 such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various embodiments of the permafrost subgrade deformation prediction method can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0133] The above-mentioned embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for predicting deformation of permafrost roadbed, characterized in that: include: Obtain historical deformation data of frozen soil roadbed and corresponding multiple variables affecting the deformation of frozen soil roadbed; For any of the variables, variational modal decomposition is performed to obtain multiple modal components of the variable; For multiple modal components of any of the variables, based on the kernel principal component analysis method, multiple principal components are screened to remove redundancy, wherein any of the principal components is a linear combination of the modal components; For any principal component, determining the correlation between the principal component and the historical deformation data of the frozen soil roadbed; The principal components with correlation greater than the threshold among the principal components are taken as the key factor data set; The key factor data set and the frozen soil roadbed historical deformation data are used as training sets to perform neural network training to obtain a frozen soil roadbed deformation prediction model; Based on the frozen soil roadbed deformation prediction model, the frozen soil roadbed deformation is predicted.

2. The method for predicting deformation of permafrost roadbed according to claim 1, characterized in that: The types of modal components include periodic terms, trend terms, and residual terms.

3. The method for predicting deformation of permafrost roadbed according to claim 1, characterized in that: After obtaining the multiple modal components of the variable, the method further includes: For each modal component, based on maximum and minimum normalization, a linear transformation is performed on the modal component to obtain a normalized component; The normalized component is used as the modal component.

4. The method for predicting deformation of permafrost roadbed according to claim 1, characterized in that: The frozen soil roadbed deformation prediction model includes a fusion algorithm; the fusion algorithm is formed by combining multiple algorithms.

5. The method for predicting deformation of permafrost roadbed according to claim 4, characterized in that: The frozen soil roadbed deformation prediction model includes a long short-term memory network and a gradient boosting decision tree.

6. The method for predicting deformation of permafrost roadbed according to claim 5, characterized in that: Based on the frozen soil roadbed deformation prediction model, the frozen soil roadbed deformation prediction includes: Obtain data on key factors affecting frozen soil roadbed deformation; The long short-term memory network extracts time-dependent features from the key factor data and outputs them to the gradient boosting decision tree; The gradient boosting decision tree utilizes the static distribution information of the time-dependent features to perform regression or classification tasks and generate frozen soil roadbed deformation prediction results.

7. The method for predicting deformation of permafrost roadbed according to claim 1, characterized in that: The plurality of modal components of any of the variables, based on the kernel principal component analysis method, screen out the plurality of principal components after removing redundancy, including: constructing an input matrix based on a plurality of modal components of any of the variables; According to the input matrix, a plurality of principal components are obtained based on a kernel principal component analysis method, wherein any of the principal components is a linear combination of each modal component; For any principal component, determine the contribution rate of the principal component; Arrange the contribution rates of the principal components in descending order, and select the principal components whose contribution rates are greater than the threshold as the principal components after removing redundancy.

8. The method for predicting deformation of permafrost roadbed according to claim 1, characterized in that: The step of determining the correlation between any principal component and the historical deformation data of the frozen soil roadbed comprises: The correlation is determined based on the following formula: in, Indicates relevance; Indicates The value of the principal component; Indicates The average of the principal components; Indicates deformation value; Indicates the average deformation value.

9. A permafrost roadbed deformation prediction device, characterized in that: include: An acquisition module is used to acquire historical deformation data of frozen soil roadbed and corresponding multiple variables affecting the deformation of frozen soil roadbed; A decomposition module, used for performing variational modal decomposition on any of the variables to obtain multiple modal components of the variable; A screening module, for screening multiple modal components of any of the variables based on a kernel principal component analysis method to obtain multiple principal components after redundancy removal, wherein any of the principal components is a linear combination of the modal components; A correlation determination module, for determining, for any principal component, the correlation between the principal component and the historical deformation data of the frozen soil roadbed; A judgment module is used to take the principal components whose correlation is greater than a threshold value among the principal components as a key factor data set; A training module, used to use the key factor data set and the frozen soil roadbed historical deformation data as training sets to perform neural network training to obtain a frozen soil roadbed deformation prediction model; The prediction module is used to predict the deformation of the frozen soil roadbed based on the frozen soil roadbed deformation prediction model.

10. 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 processor executes the computer program, the steps of the method for predicting deformation of permafrost roadbed as described in any one of claims 1 to 8 are implemented.

Images