Frost development prediction method and device, electronic equipment and storage medium

Abstract

This application provides a method, device, electronic equipment, and storage medium for predicting permafrost development, belonging to the field of soil analysis technology. The method includes: acquiring climate data of a target area; recursively and adaptively dividing the target area based on the climate data; calculating the climate data diffusion intensity of each target sub-region; inputting the climate data and diffusion intensity corresponding to the target sub-region into a pre-trained climate prediction model, outputting predicted climate data for multiple time steps, wherein the pre-trained climate prediction model includes multiple LSTM modules, each LSTM module including a physical constraint forgetting gate related to the climate data diffusion intensity; and inputting the predicted climate data for each time step into a pre-trained classification model, outputting the predicted permafrost state of the corresponding target sub-region. This method deeply integrates data-driven approaches with physical mechanisms, significantly improving the accuracy and reliability of permafrost development state prediction under complex environments through the synergistic effect of adaptive partitioning and physical constraints.

Description

Technical Field

[0001] This application belongs to the field of soil analysis technology, and more specifically, relates to a method and apparatus for predicting permafrost development, electronic equipment, and storage medium. Background Technology

[0002] Permafrost refers to a type of soil or rock that has formed due to the freezing of moisture under prolonged low temperatures. It typically occurs in cold regions, such as the Arctic, mountainous areas, or high-latitude regions, where temperatures and humidity are low, and the moisture in the soil is mostly in an ice-like state.

[0003] Forecasting permafrost development is crucial because it directly impacts the safety and stability of engineering projects in cold regions. With climate change, permafrost thawing can loosen soil structures, threatening the foundation stability of infrastructure construction and operation, especially large-scale projects like subways. Accurately predicting permafrost trends helps engineers take appropriate design and reinforcement measures in advance, ensuring foundation stability during construction and operation, and avoiding potential safety risks and economic losses due to permafrost issues.

[0004] However, existing permafrost prediction methods typically employ fixed or static regional divisions, which are difficult to adapt to the spatial heterogeneity of climate data. Furthermore, traditional predictions are usually based solely on data-driven approaches to forecast climate data, resulting in insufficient accuracy in predicting the state of permafrost development based on predicted climate data. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, electronic equipment, and storage medium for predicting permafrost development, in order to solve the problems that existing permafrost prediction methods usually adopt fixed or static regional division methods, which are difficult to adapt to the spatial heterogeneity of climate data, and traditional predictions are usually based solely on data-driven methods to predict climate data, which leads to insufficient accuracy in predicting the state of permafrost development based on predicted climate data.

[0006] A first aspect of this application provides a method for predicting permafrost development, the method comprising: Obtain climate data for the target area; Based on climate data, the target area is recursively and adaptively divided to obtain multiple target sub-regions; Calculate the diffusion intensity of climate data for each target sub-region; The climate data and climate data diffusion intensity corresponding to the target sub-region are input into the pre-trained climate prediction model, and the predicted climate data at multiple time steps are output. The pre-trained climate prediction model includes multiple LSTM modules, and each LSTM module includes a physical constraint forget gate related to the climate data diffusion intensity. The predicted climate data at each time step is input into the pre-trained classification model, which outputs the predicted permafrost state of the corresponding target sub-region.

[0007] A second aspect of this application provides a permafrost development prediction device, the device comprising: The acquisition module is used to acquire climate data for the target area; The partitioning module is used to recursively and adaptively partition the target area based on climate data, resulting in multiple target sub-regions. The calculation module is used to calculate the diffusion intensity of climate data for each target sub-region; The first output module is used to input the climate data and climate data diffusion intensity corresponding to the target sub-region into the pre-trained climate prediction model and output the predicted climate data for multiple time steps. The pre-trained climate prediction model includes multiple LSTM modules, and each LSTM module includes a physical constraint forget gate related to the climate data diffusion intensity. The second output module is used to input the predicted climate data at each time step into the pre-trained classification model and output the predicted permafrost state of the corresponding target sub-region.

[0008] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the permafrost development prediction method described above.

[0009] In a fourth aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described permafrost development prediction method.

[0010] The beneficial effects of the permafrost development prediction method and apparatus, electronic equipment, and storage medium provided in this application are as follows: In this embodiment, spatial heterogeneity is effectively captured by acquiring climate data and performing adaptive regional division based on it. The impact of key physical processes on the surrounding areas is quantified by calculating the diffusion intensity of climate data between target sub-regions. Furthermore, the diffusion intensity is incorporated as a physical constraint into the forgetting gate mechanism of LSTM, guiding the model to follow the basic laws of energy and matter propagation during learning, thereby generating a climate prediction sequence that better reflects physical reality. Finally, permafrost status is classified based on this high-precision climate sequence. The entire process deeply integrates data-driven approaches with physical mechanisms, significantly improving the accuracy and reliability of permafrost development state prediction in complex environments through the synergistic effect of adaptive division and physical constraints. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a method for predicting permafrost development according to an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a pre-trained climate prediction model provided in an embodiment of this application; Figure 3 A structural block diagram of a permafrost development prediction device provided in one embodiment of this application; Figure 4 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] It is understood that in the embodiments of this application, data such as user information are involved. When the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with relevant laws, regulations and standards.

[0015] It should be noted that the terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.

[0016] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0018] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for predicting permafrost development according to an embodiment of this application.

[0019] like Figure 1 As shown, the permafrost development prediction method provided in this application embodiment may include: S1. Obtain climate data for the target area.

[0020] The target area refers to the geographical space within which permafrost development prediction needs to be conducted. Climate data refers to the key meteorological and environmental variables affecting the permafrost state within this area. By focusing on specific target areas and acquiring their multi-dimensional climate data, a comprehensive and reliable data foundation is provided for subsequent accurate analysis and modeling.

[0021] In one possible implementation, climate data includes surface temperature, precipitation, humidity, wind direction, wind speed, and air pressure.

[0022] Understandably, climate data, which includes multi-dimensional elements, can more comprehensively depict the energy and water exchange processes that affect permafrost development, thereby providing models with more accurate and physically complete input information and significantly improving the reliability of predictions.

[0023] S2. Based on climate data, the target area is recursively and adaptively divided to obtain multiple target sub-regions.

[0024] The recursive adaptive partitioning method refers to dividing a target region into multiple sub-regions with relatively uniform internal climate conditions but significant variations at the boundaries, based on the spatial distribution characteristics of climate data and using an iterative algorithm. This method does not pre-fix the grid but rather allows the partitioning boundaries to adapt to the data itself. This approach can automatically identify and finely characterize the spatial heterogeneity of climate factors, providing more accurate spatial units for subsequent targeted predictions. It overcomes the shortcomings of fixed grids, which ignore local features and lead to large deviations and low accuracy in permafrost development predictions.

[0025] In one possible implementation, the target area is recursively and adaptively divided based on climate data to obtain multiple target sub-regions, specifically including: The target area is divided into multiple uniform grids.

[0026] Calculate the global difference entropy, which describes the degree of disorder in climate differences among all grid pairs.

[0027] A grid pair refers to two grids within the target area.

[0028] The formula for calculating global difference entropy is as follows: .

[0029] .

[0030] in, Represents global difference entropy. They represent the first i The grid and the first j Climate data vectors of grids, Indicates the total number of grid cells. express and Normalized parameters of climate differences between them Represents the logarithmic function. Represents the natural index. Indicates the first i The grid and the first j Gaussian kernel values ​​for each grid cell Represents the Gaussian kernel scale parameter. The value represents the normalization parameter, which is the sum of the Gaussian kernel values ​​for all grid pairs. It represents a very small positive number.

[0031] Optionally, the smallest positive number can be taken as... The purpose is to ensure that logarithmic operations are meaningful and stable. The Gaussian kernel scale parameter is used to control the sensitivity to differences in climate characteristics, and can be taken as the median of the Euclidean distance across all grids.

[0032] It should be noted that the global difference entropy, by combining the Gaussian kernel function with information entropy, transforms complex spatial climate differences into a stable and quantifiable overall disorder index. This can sensitively and robustly characterize the degree of heterogeneity of climate features within a region, providing an objective and reliable decision-making basis for subsequent adaptive classification and effectively avoiding biases caused by subjective threshold settings.

[0033] By combining the geographical scale and climate data complexity of the target area, an upper limit for the entropy value of the target area is determined.

[0034] The formula for calculating the upper limit of entropy is as follows: .

[0035] in, Indicates the area of ​​the target region. This represents the dimension of climate data for the target region, i.e., the complexity of the climate data. This indicates the upper limit of the entropy value of the target region. This represents the geographic scale weighting factor.

[0036] The geographic scale weighting factor is used to adjust the entropy calculation based on the geographic characteristics of the target area, taking into account the area, shape, and climate complexity of the region to avoid excessive influence of large areas on the entropy value, thus making the calculation results more consistent with actual geographic and climatic differences. Specifically, the value can range from 0.1 / meter to 1 / meter.

[0037] It should be noted that by introducing a theoretical entropy upper limit based on area and data dimensions, a dynamic and objective stopping criterion is provided for the recursive partitioning. This avoids the bias caused by subjectively setting fixed thresholds, ensuring that the partitioning results can fully characterize spatial heterogeneity without generating meaningless oversegmentation, thus achieving an optimal balance between computational efficiency and geophysical significance. (The square root of the target area is then calculated.) This makes the impact of the target area area on the global differential entropy smoother and more reasonable, forming a geographical scale that affects the upper limit of the entropy value, avoiding the problem of excessive entropy growth when the area increases, thus making the calculation results more consistent with reality.

[0038] The ratio of the global difference entropy to the upper limit of the target region entropy value is calculated to obtain the partitioning decision coefficient.

[0039] If the decision coefficient is greater than 1, proceed to the next step; otherwise, treat the target region as the target sub-region and proceed to the final step, which is to output the target sub-region.

[0040] The Gaussian kernel value between grid pairs is used as the grid pair similarity output, and the similarity of each grid pair is used to form a grid pair similarity matrix.

[0041] The Laplacian matrix is ​​calculated by combining the grid with the similarity matrix.

[0042] The formula for calculating the Laplace matrix is ​​as follows: .

[0043] in, Represents the Laplace matrix, Degree matrix, Represents the similarity matrix. Represents the identity matrix.

[0044] The degree matrix is ​​a diagonal matrix whose diagonal elements are equal to the sum of all elements in the corresponding row (or column) of the similarity matrix. It describes the total strength of the connection between each node (i.e., the grid) and other nodes in the graph. This process can eliminate the influence of differences in the connectivity of nodes themselves on the segmentation results by normalizing the similarity matrix, ensuring that the spectral clustering algorithm can more fairly and stably divide regions based on the similarity between nodes, thereby obtaining sub-region boundaries with more reasonable physical meaning.

[0045] Solve for the Fiedler vector of the Laplacian matrix, where the Fiedler vector is specifically the eigenvector corresponding to the second smallest eigenvalue of the Laplacian matrix.

[0046] The second smallest eigenvalue is the second smallest eigenvalue of the Laplace matrix. More formally, it is described as the eigenvector corresponding to the (q-1)th eigenvalue of the Laplace matrix, where the (q-1)th eigenvalue is less than the qth eigenvalue, and its qth eigenvalue is the smallest eigenvalue of the Laplace matrix.

[0047] Based on the sign of each element in the Fiedler vector, the target region is divided into the first region and the second region.

[0048] Each grid corresponds to a component in the Fiedler vector. The component has a positive or negative sign, and the regions corresponding to the positive and negative signs are the first and second regions, respectively.

[0049] Specifically, this process utilizes the principle of spectral clustering to partition regions by solving for the Fiedler vector of the Laplacian matrix. The core idea is that the Fiedler vector, as the eigenvector corresponding to the second smallest eigenvalue, essentially represents the optimal relaxation solution to the graph cut problem through the sign of each component. This sign indicates the best partitioning method that results in tight connectivity within sub-regions and sparse connectivity between regions. Specifically, a positive sign corresponds to the first region, and a negative sign corresponds to the second region. Its advantage lies in being a globally optimized partitioning method based on the similarity structure of the data itself. It can automatically and objectively identify the natural spatial boundaries of climate characteristics, ensuring high climate uniformity within each partitioned sub-region, thus laying an ideal spatial unit foundation for subsequent accurate predictions.

[0050] Using the first and second regions as target regions respectively, return to step one, that is, return to dividing the target region into multiple uniform grids.

[0051] Output the target sub-region.

[0052] Specifically, this process achieves adaptive region partitioning through a recursive algorithm. It uses information entropy to quantify and assess the spatial heterogeneity of climate data within a region, and decides whether to further segment by comparing it with theoretical entropy limits. When there are significant climate differences within a region, the algorithm uses spectral clustering to find the optimal segmentation boundary based on the similarity matrix of climate features between grids, dividing the region into two more similar sub-regions, and recursively applying this process. Its advantage lies in being entirely data-driven, automatically identifying and following the natural distribution patterns of climate factors for refined segmentation, thus ensuring a high degree of consistency in climate conditions within each final sub-region. This effectively overcomes the neglect of local features by manual or fixed partitioning methods, providing more spatially representative analytical units for subsequent permafrost prediction, fundamentally improving the model's input quality and prediction accuracy.

[0053] S3. Calculate the diffusion intensity of climate data for each target sub-region.

[0054] In one possible implementation, the climate data diffusion intensity of each target sub-region is calculated, specifically as follows: The diffusion intensity of climate data is calculated using the Laplace operator.

[0055] The specific formula for calculating the diffusion intensity of climate data is as follows:

[0056] in, Represents the Laplace operator. Indicates based on t Climate data at -1 hour Definite t Intensity of climate data diffusion at any given time.

[0057] Climate data diffusion intensity is a measure of the spread and extent of climate change within a region. The Laplace operator reflects the rate of change of climate data spatially, thus determining the diffusion intensity. Calculating climate data diffusion intensity using the Laplace operator effectively captures the spatial variation characteristics of climate data, helping to more accurately analyze the speed and intensity of climate change in different regions. This method can improve the spatial resolution of climate data predictions and enhance the accuracy of prediction results.

[0058] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of a pre-trained climate prediction model provided in an embodiment of this application.

[0059] Figure 2 In the middle, mark k This represents the number of LSTM modules that correspond to the dimensionality of the climate data.

[0060] S4. Input the climate data and climate data diffusion intensity corresponding to the target sub-region into the pre-trained climate prediction model, and output the predicted climate data for multiple time steps.

[0061] The pre-trained climate prediction model includes multiple LSTM modules, each of which includes a physical constraint forget gate related to the intensity of climate data diffusion.

[0062] The physical constraint forgetting gate refers to introducing the diffusion intensity of climate data calculated in the previous step as a physical constraint into the standard forgetting gate structure of the LSTM network. This ensures that the decision to retain or forget historical information is guided not only by statistical data patterns but also by the physical law of diffusion intensity of climate data in neighboring regions. By embedding diffusion intensity, reflecting spatial physical processes, as a core constraint into the LSTM forgetting mechanism, the physical rationality and accuracy of climate sequence predictions are significantly improved. This fusion ensures that the model follows the basic principles of energy and matter diffusion in time series learning, thereby generating more reliable multi-step prediction results that are more in line with natural laws and improving the accuracy of subsequent permafrost state classification.

[0063] In one possible implementation, the number of LSTM modules is the same as the number of dimensions of the climate data, with each LSTM module corresponding to one dimension of climate data.

[0064] The output layers of each LSTM module are connected sequentially.

[0065] Each LSTM module includes an input layer, input gates, a physical constraint forget gate, an output gate, a memory cell, and an output layer. The input layer is connected to the memory cell via the input gates. The physical constraint forget gate is connected to the memory cell. The memory cell is connected to the output layer via the output gate.

[0066] Specifically, the pre-trained climate prediction model adopts a multi-module parallel structure. Each LSTM module focuses on processing the temporal features of a single climate dimension, and then multivariate information is fused through the cascading of output layers. By introducing a physically constrained forget gate, prior knowledge such as climate diffusion intensity is embedded into the gating mechanism, enabling the model to follow the natural laws of energy and matter propagation during time series learning, thereby generating more physically reasonable and more accurate multi-step climate prediction results.

[0067] In one possible implementation, the pre-trained classification model is specifically a random forest classification model. The predicted climate data at each time step is input into the pre-trained classification model, which outputs the predicted permafrost state for the corresponding target sub-region, specifically including: Climate data is fed into the input gate through the input layer to obtain the input gate output and candidate states.

[0068] The specific formulas for calculating the input gate output and candidate states are as follows: .

[0069] .

[0070] in, and They represent t Input gate output and candidate state at time t. and These represent the input gate weights and input gate biases, respectively. and These represent the candidate state adjustment weight and the candidate state adjustment bias, respectively. This represents the sigmoid activation function. express t -1 is the hidden state. t Output gate output at time -1 express t Input data at any time t Real-time climate data.

[0071] Understandably, real-time input data, i.e., climate data, is often unavailable during the forecasting process. In such cases, forecasts are typically made based solely on the output data from the previous moment.

[0072] Climate data input and climate data diffusion intensity are fed into a physically constrained forget gate to obtain a physically constrained forget gate output.

[0073] The specific formula for calculating the output of the physical constraint forget gate is as follows: .

[0074]

[0075] in, express t The physical constraints of the forget gate output at any given time. and These represent the physical constraint forget gate weight and the physical constraint forget gate bias, respectively. This represents the learnable physical constraint sensitivity coefficient. This indicates element-wise multiplication. This represents physical constraint terms.

[0076] It should be noted that this physically constrained forget gate generates a dynamic adjustment term by incorporating the norm of climate diffusion intensity into the sigmoid function and multiplying it element-wise with the output of a traditional forget gate. This allows for dynamic filtering of historical memories based on the physical laws governing the spatial propagation of energy or matter. This enables the model to spontaneously follow physical processes during time-series learning. When inter-regional climate interactions are strong, the dependence on local historical states is automatically reduced, effectively improving the physical rationality and spatiotemporal accuracy of multi-step climate predictions and avoiding biases that might arise from purely data-driven approaches that violate natural laws.

[0077] Climate data is input into the output gate, and the output gate outputs the data.

[0078] The specific formula for calculating the output of the output gate is as follows: .

[0079] in, express t Output gate output at all times, and These represent the output gate weight and the output gate bias, respectively.

[0080] Update the memory cell state based on the input gate output, candidate state, and physical constraint forget gate output.

[0081] The specific formula for updating the state of a memory cell is as follows: .

[0082] in, and They represent t Time and t The state of the memory cell at time -1.

[0083] Based on the updated memory cell state and output gate output, the predicted climate data is obtained and then output through the output layer.

[0084] The specific formula for calculating predicted climate data is as follows: .

[0085] in, express t Real-time climate data forecasting express Activation function.

[0086] Update the climate data using the predicted climate data, return to step one, and continue until the predicted climate data for the preset time step is obtained.

[0087] It should be noted that those skilled in the art can set the size of the preset time step according to actual needs, and this invention does not limit this.

[0088] Specifically, this method uses an improved LSTM module for iterative prediction, transforming the intensity of climate data diffusion reflecting space physical processes into a learnable physical constraint term. This constraint is then multiplied element-wise with the output of a standard forget gate, dynamically adjusting the retention of historical memory. This mechanism allows the model to spontaneously filter information based on the intensity of energy or matter diffusion in space when learning time-series patterns: when the diffusion effect is strong, the physical constraint term guides the model to forget more isolated local historical states and pay more attention to the synergistic effects from surrounding areas. Conversely, it relies more on local historical continuity. By embedding physical priors into the data-driven model, the entire prediction process not only learns statistical regularities but also inherently follows the dynamic equilibrium principle of nature, thus significantly improving the physical consistency and spatiotemporal accuracy of multi-step climate predictions, ultimately providing more reliable climate data sequence input for permafrost state classification.

[0089] Optionally, the specific training process of this pre-trained climate prediction model is as follows: First, prepare a training dataset containing historical climate sequences and their corresponding diffusion intensities. In each iteration, the climate data is input into the corresponding LSTM module according to its dimensions. During forward propagation, the physical constraint forget gate of each module dynamically generates constraint terms based on the input climate diffusion intensity norm and combines them with the traditional forget gate signal to regulate the update of the memory unit. Subsequently, the loss (such as mean squared error) between the multi-step predictions output by the model and the actual future climate data is calculated. Finally, through the backpropagation algorithm, the various weight parameters of the LSTM module and the physical constraint sensitivity coefficient are optimized simultaneously, so that the model learns how to effectively utilize physical laws to constrain temporal memory while minimizing the prediction error, thereby achieving the training goal of deep integration of data-driven and physical mechanisms.

[0090] S5. Input the predicted climate data at each time step into the pre-trained classification model and output the predicted permafrost state of the corresponding target sub-region.

[0091] In one possible implementation, the pre-trained classification model is specifically a random forest classification model. The predicted climate data at each time step is input into the pre-trained classification model, which outputs the predicted permafrost state for the corresponding target sub-region, specifically including: The predicted climate data at each time step are input into the random forest classification model.

[0092] Output the predicted permafrost state for the corresponding target sub-region, where the predicted permafrost state includes both frozen and unfrozen states.

[0093] Furthermore, the predicted frozen soil state is spliced ​​together according to the time step sequence to obtain the predicted frozen soil state time series.

[0094] Among them, the random forest classification model is an ensemble learning method that classifies data by constructing a large number of decision trees and combining their voting results. This model has high training efficiency, is insensitive to data noise, and can effectively handle the complex nonlinear relationship between climate variables and permafrost conditions, ensuring the stability and reliability of the final state prediction results.

[0095] The random forest classification model consists of a set of independent decision trees, each of which is a complete classifier with a root node, internal nodes, and leaf nodes. The root and internal nodes partition the data based on a threshold of a certain dimension of the input features (i.e., the predicted climate data at each time step), while the leaf nodes store the final classification result (frozen or unfrozen state). The model's final output is generated by aggregating the classification results from all decision trees using a majority voting mechanism.

[0096] The specific training process of this random forest classification model is as follows: First, multiple sample subsets are extracted from the training dataset with replacement using a bootstrap sampling method, and a decision tree is built in parallel for each subset. During the construction of each tree, when a node needs to be split, the algorithm randomly selects a feature subset from all climate features, choosing the feature that minimizes Gini impurity or information entropy and its threshold as the splitting criterion. This process is repeated recursively until a preset termination condition is met (such as too few node samples or reaching the maximum depth). Finally, all the grown decision trees constitute the random forest model. Through this parallel ensemble method that introduces randomness, the model effectively reduces variance and enhances generalization ability.

[0097] In one possible implementation, it also includes: The pre-trained climate prediction model and the pre-trained classification model are retrained at preset intervals.

[0098] It should be noted that those skilled in the art can set the preset duration according to actual needs, and this application embodiment does not limit it.

[0099] In practical applications, this method first uses multidimensional climate data to drive recursive adaptive partitioning, accurately dividing the target region into sub-units with homogeneous internal climates, effectively overcoming the neglect of spatial heterogeneity by fixed grids. Next, it calculates the diffusion intensity of climate data in each sub-region and embeds it as a physical constraint into the forgetting gate mechanism of an improved LSTM prediction model. This ensures that climate sequence predictions not only follow statistical data patterns but also inherently conform to the natural principles of energy and matter diffusion, thus generating physically reasonable and more accurate multi-step climate predictions. Finally, a trained random forest classification model is used to discriminate these predicted climate sequences, stably outputting the permafrost state of each sub-region over time. The entire process deeply integrates data-driven and physical mechanisms, comprehensively improving the accuracy and reliability of predictions from spatial foundations to temporal evolution through the synergy of adaptive partitioning and physically constrained prediction.

[0100] In this embodiment, spatial heterogeneity is effectively captured by acquiring climate data and performing adaptive regional division based on it. The impact of key physical processes on the surrounding areas is quantified by calculating the diffusion intensity of climate data between target sub-regions. Furthermore, the diffusion intensity is incorporated as a physical constraint into the forgetting gate mechanism of LSTM, guiding the model to follow the fundamental laws of energy and matter propagation during learning, thereby generating a climate prediction sequence that better reflects physical reality. Finally, permafrost status is classified based on this high-precision climate sequence. The entire process deeply integrates data-driven approaches with physical mechanisms, significantly improving the accuracy and reliability of permafrost development state prediction under complex environments through the synergistic effect of adaptive division and physical constraints.

[0101] Based on the same inventive concept, this application also provides a permafrost development prediction device for implementing the permafrost development prediction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more permafrost development prediction device embodiments provided below can be found in the limitations of the permafrost development prediction method described above, and will not be repeated here.

[0102] This application provides a permafrost development prediction device, such as... Figure 3 As shown, the permafrost development prediction device 20 includes: The acquisition module 201 is used to acquire climate data for the target area.

[0103] The partitioning module 202 is used to perform adaptive partitioning of the target area in a recursive form based on climate data, resulting in multiple target sub-regions.

[0104] The calculation module 203 is used to calculate the diffusion intensity of climate data for each target sub-region.

[0105] The first output module 204 is used to input the climate data and climate data diffusion intensity corresponding to the target sub-region into the pre-trained climate prediction model and output the predicted climate data for multiple time steps. The pre-trained climate prediction model includes multiple LSTM modules, and each LSTM module includes a physical constraint forget gate related to the climate data diffusion intensity.

[0106] The second output module 205 is used to input the predicted climate data at each time step into the pre-trained classification model and output the predicted permafrost state of the corresponding target sub-region.

[0107] See Figure 4 , Figure 4 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 4 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of each module / unit in the above-described device embodiments, for example... Figure 3 The functions of the acquisition module 201, the division module 202, the calculation module 203, the first output module 204, and the second output module 205 are shown.

[0108] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may 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 any conventional processor.

[0109] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0110] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0111] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation method described in the permafrost development prediction method provided in the embodiments of this application, or they can execute the implementation method of the electronic device described in the embodiments of this application, which will not be repeated here.

[0112] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0113] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., provided on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0114] Those skilled in the art will recognize that the modules / units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0115] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0116] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules, units, or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or modules / units, or it may be an electrical, mechanical, or other form of connection.

[0117] The modules / units described as separate components may or may not be physically separate. Similarly, the components shown as modules / units may or may not be physical modules / units; they may be located in one place or distributed across multiple network modules / units. Some or all of the modules / units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0118] Furthermore, the functional modules / units in the various embodiments of this application can be integrated into one processing module / unit, or each module / unit can exist physically separately, or two or more modules / units can be integrated into one module / unit. The integrated modules / units described above can be implemented in hardware or in the form of software functional modules / units.

[0119] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting permafrost development, characterized in that, The method includes: Obtain climate data for the target area; Based on the climate data, the target area is recursively and adaptively divided to obtain multiple target sub-regions; Calculate the climate data diffusion intensity for each of the target sub-regions; The climate data corresponding to the target sub-region and the climate data diffusion intensity are input into the pre-trained climate prediction model, and the predicted climate data for multiple time steps are output. The pre-trained climate prediction model includes multiple LSTM modules, and each LSTM module includes a physical constraint forget gate related to the climate data diffusion intensity. The predicted climate data for each of the time steps are input into the pre-trained classification model, which outputs the predicted permafrost state for the corresponding target sub-region.

2. The method for predicting permafrost development as described in claim 1, characterized in that, The recursive adaptive partitioning of the target region based on the climate data to obtain multiple target sub-regions specifically includes: The target region is divided into multiple uniform grids; Calculate the global difference entropy that describes the degree of disorder in climate differences among all grid pairs; Based on the geographical scale and climate data complexity of the target area, determine the upper limit of the entropy value of the target area; The ratio of the global difference entropy to the upper limit of the target region entropy value is calculated to obtain the partitioning decision coefficient; If the division decision coefficient is greater than 1, proceed to the next step; otherwise, treat the target region as the target sub-region and proceed to the final step. The Gaussian kernel value between grid pairs is used as the grid pair similarity output, and the similarity of each grid pair is used to form a grid pair similarity matrix; Calculate the Laplacian matrix by combining the grid with the similarity matrix; Solve for the Fiedler vector of the Laplacian matrix, wherein the Fiedler vector is specifically the eigenvector corresponding to the second smallest eigenvalue of the Laplacian matrix; Based on the sign of each element in the Fiedler vector, the target region is divided into a first region and a second region; Using the first region and the second region as the target regions respectively, return to step one; Output the target sub-region.

3. The permafrost development prediction method as described in claim 1, characterized in that, The calculation of the climate data diffusion intensity for each of the target sub-regions specifically includes: The diffusion intensity of the climate data is calculated using the Laplace operator.

4. The method for predicting permafrost development as described in claim 1, characterized in that, The number of LSTM modules is the same as the number of dimensions of the climate data, and each LSTM module corresponds to one dimension of climate data; The output layers of each of the LSTM modules are connected sequentially; Each LSTM module includes an input layer, an input gate, a physical constraint forget gate, an output gate, a memory unit, and an output layer; the input layer is connected to the memory unit through the input gate; the physical constraint forget gate is connected to the memory unit; and the memory unit is connected to the output layer through the output gate.

5. The permafrost development prediction method as described in claim 4, characterized in that, The step of inputting the climate data corresponding to the target sub-region and the climate data diffusion intensity into the pre-trained climate prediction model, and outputting predicted climate data for multiple time steps, specifically includes: The climate data is input to the input gate through the input layer to obtain the input gate output and candidate states; The climate data and the climate data diffusion intensity are input into the physical constraint forget gate to obtain the physical constraint forget gate output. The climate data is input into the output gate, and the output gate output is obtained; Update the memory cell state based on the input gate output, the candidate state, and the physical constraint forget gate output; The predicted climate data is obtained based on the updated memory cell state and the output gate output, and the predicted climate data is output through the output layer. The predicted climate data is updated using the predicted climate data, and the process returns to step one until the predicted climate data for the preset time step is obtained.

6. The method for predicting permafrost development as described in claim 1, characterized in that, The pre-trained classification model is specifically a random forest classification model; the step of inputting the predicted climate data at each time step into the pre-trained classification model and outputting the predicted permafrost state of the corresponding target sub-region specifically includes: The predicted climate data for each of the time steps are input into the random forest classification model; Output the predicted permafrost state of the corresponding target sub-region, wherein the predicted permafrost state includes frozen state and unfrozen state.

7. The method for predicting permafrost development as described in claim 1, characterized in that, The method further includes: The pre-trained climate prediction model and the pre-trained classification model are retrained at preset intervals.

8. A permafrost development prediction device, characterized in that, The device includes: The acquisition module is used to acquire climate data for the target area; The partitioning module is used to recursively and adaptively partition the target area based on the climate data to obtain multiple target sub-regions. The calculation module is used to calculate the climate data diffusion intensity of each of the target sub-regions; The first output module is used to input the climate data corresponding to the target sub-region and the climate data diffusion intensity into the pre-trained climate prediction model and output the predicted climate data for multiple time steps. The pre-trained climate prediction model includes multiple LSTM modules, and each LSTM module includes a physical constraint forget gate related to the climate data diffusion intensity. The second output module is used to input the predicted climate data of each time step into the pre-trained classification model and output the predicted permafrost state of the corresponding target sub-region.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

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

Images