Soil humidity prediction method of attention coding and decoding LSTM (Long Short Term Memory) model based on physical process

Through attention encoding and decoding of the LSTM model based on physical processes, combined with the intermediate variables of the HBV hydrological physical model, the problem of spatial and temporal variability and physical laws integration in soil moisture prediction is solved, and efficient and accurate soil moisture prediction is achieved, especially at the extreme Climate areas show excellent performance.

CN118940160AActive Publication Date: 2024-11-12CHANGCHUN NORMAL UNIV

Patent Information

Application Number
CN202411153495.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-21
Publication Date
2024-11-12
Estimated Expiration
2044-08-21

AI Technical Summary

Technical Problem

Existing soil moisture prediction models are difficult to effectively deal with the spatiotemporal variability of soil characteristics and integrate physical processes and data-driven models to improve prediction accuracy. The lack of physical mechanisms in deep learning models leads to the prediction results in inconsistent with actual physical laws.

Method used

The LSTM model is decoded based on physical processes. By obtaining the surface soil moisture and its influencing elements at historical moments, input the HBV hydrological physical model to obtain intermediate variable data, perform correlation analysis and select key feature variables, and combine attention coding. Decode the LSTM model for training to improve the accuracy of soil moisture prediction.

Benefits of technology

The soil moisture prediction combined with deep learning and physical models is realized, which improves the flexibility and accuracy of prediction, especially in soil moisture prediction in extreme climate areas, which shows performance better than traditional deep learning models, demonstrating deep learning and The strong potential of physical models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940160A_ABST
    Figure CN118940160A_ABST
Patent Text Reader

Abstract

The invention provides a soil humidity prediction method based on an attention coding and decoding LSTM model of a physical process, and the method comprises the steps: obtaining surface soil humidity values at all historical moments and influence factors of the surface soil humidity values, and the influence factors comprise an atmosphere forced variable, a land surface variable and a static variable; inputting part of the influence elements into the HBV hydrological physical model and operating the HBV hydrological physical model to obtain intermediate variable data; performing correlation analysis on the intermediate variable data and the surface soil humidity, and selecting an intermediate characteristic variable combination most relevant to the surface soil humidity; the influence factors and the intermediate characteristic variables are combined to serve as input data, a preset attention-based coding and decoding LSTM model is trained, so that the model reaches preset precision, and a trained soil humidity model is obtained; and predicting the surface soil humidity at the future moment through the trained soil humidity model. According to the method, the powerful potential of combination of deep learning and a physical model is shown, and a new direction is provided for an efficient and accurate hydrological model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of soil moisture prediction, and in particular to a soil moisture prediction method based on a physical process-based attention encoding and decoding LSTM model. Background Art

[0002] Soil is a natural material whose properties may vary significantly in space and time. Traditional machine learning models are usually not capable of handling time dependencies and capturing long-term dependencies when processing soil moisture time series data, while deep learning models are specifically designed to handle these challenges and can better predict future states. Deep learning models are able to learn complex nonlinear functions and transform low-level information into high-level features, thereby enhancing the representation of raw data. They are flexible and efficient in processing data and feature extraction, and can automatically handle missing data and noise, while extracting effective features from multiple data sources, such as meteorological data, soil types, etc., further improving prediction accuracy. These models show excellent generalization capabilities in soil moisture prediction and are applicable to different regions, seasons, and soil types.

[0003] In hydrology, the long short-term memory (LSTM) architecture is particularly popular and has shown great performance advantages in soil moisture prediction, representing a "performance leap". However, although the LSTM model has shown high accuracy and generalization ability in soil moisture simulation, deep learning lacks physical mechanisms, which may lead to predictions that do not conform to physical laws and false predictions, limiting its application.

[0004] Physical hydrological models typically allow for spatial variability in different parameters as well as dynamics over time, which is critical for capturing hydrological peaks and valleys caused by geographic or seasonal variations. However, they typically require significant computational resources and may not accurately capture complex nonlinear relationships. In contrast, deep learning models learn complex patterns and relationships in a data-driven manner, reducing computational requirements. While deep learning models may lack an understanding of physical laws, they excel in capturing complex nonlinear relationships.

[0005] The challenge of current soil moisture prediction models is to effectively handle the spatiotemporal variability of soil properties and integrate physical processes with data-driven models to improve prediction accuracy. Although deep learning models, especially LSTM and its variants, perform well in processing time series and long-term dependencies, they are usually unable to directly encode physical processes, which may lead to prediction results that are inconsistent with actual physical laws. Summary of the invention

[0006] In order to solve the technical problems existing in the above background technology, the present invention provides a soil moisture prediction method based on an attention encoding and decoding LSTM model of a physical process, comprising: Obtaining the surface soil moisture value and its influencing factors at each historical moment, wherein the influencing factors include atmospheric forcing variables, land surface variables and static variables; Input some of the influencing factors into the HBV hydrological physical model and run it to obtain intermediate variable data; Performing correlation analysis between the intermediate variable data and the surface soil moisture, and selecting the intermediate characteristic variable combination most correlated with the surface soil moisture; The influencing factors and the intermediate characteristic variables are combined as input data, and a preset attention-based encoding and decoding LSTM model is trained so that the model reaches a preset accuracy, thereby obtaining a trained soil moisture model; The surface soil moisture at future times is predicted using the trained soil moisture model.

[0007] As a further illustration of the present invention, the atmospheric forcing variables include two-meter temperature, 10-meter east-west component wind speed, 10-meter north-south component wind speed, precipitation, surface air pressure, and specific humidity; The land surface variables include volume soil water layer, ground solar radiation, ground thermal radiation, soil temperature and evaporation; The static variables include soil moisture content, clay, sand, silt and elevation data from DEM.

[0008] As a further illustration of the present invention, the influencing factors input into the HBV hydrological physical model are two-meter temperature and precipitation in the atmospheric forcing variables and evaporation data in the land surface variables.

[0009] As a further illustration of the present invention, the intermediate variable data are snow melting, recondensation, soil moisture content, water recharge, water penetration into the soil, shallow aquifer and deep aquifer data.

[0010] As a further illustration of the present invention, the intermediate characteristic variable combination is soil moisture content and deep aquifer data.

[0011] As a further illustration of the present invention, the correlation analysis includes Pearson correlation coefficient calculation and random forest importance analysis.

[0012] As a further illustration of the present invention, the calculation formula of the Pearson correlation coefficient is as follows: ;

[0013] in, and denote the i-th observed value of the two variables, and They represent the average values ​​of two variables respectively, and ∑ represents the summation operation.

[0014] As a further illustration of the present invention, the influencing factors and the intermediate feature variables are combined as input data to train the preset attention-based encoding and decoding LSTM model so that the model reaches a preset accuracy, including: Dividing the influencing factors and the intermediate characteristic variable combination data into a training set, a validation set and a test set; Performing model training on the preset attention-based encoding and decoding LSTM model through the training set; Verifying the trained soil moisture model through the verification set; The verified soil moisture model is tested using the test set to ensure that the soil moisture model reaches a preset accuracy.

[0015] As a further illustration of the present invention, the training process of the attention-based encoding and decoding LSTM model specifically includes: The influencing factors and the intermediate feature variable combination are used as input data and processed by the encoder LSTM layer, so that the encoder effectively extracts the spatiotemporal feature representation in the input data and converts it into a hidden state of fixed dimension; Using a multi-head attention layer to weight the output of the encoder, highlight key features, and establish effective interactive relationships between features; The attention-weighted output data is linearly transformed through the fully connected layer to further integrate and transform the feature representation; Processing the output data through the decoder LSTM layer, using the encoder's hidden state information and previous prediction results to gradually generate an output sequence, and randomly discarding the output data; The output of the last time step is linearly transformed through the fully connected layer to obtain the final decoder output result.

[0016] As a further illustration of the present invention, the parameters of the attention-based encoding and decoding LSTM model are set as follows: the learning rate is set to 0.001, the hidden unit size is 128, the batch size is 64, the training cycle is 1000, the number of iterations is 400, the drop rate is 0.15, the head of the multi-head attention mechanism is set to 2, and the sequence length is 365 days.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects: This paper proposes an AEDLSTM (HBV) model, which is an innovative model that combines deep learning and physics-based methods to predict global soil moisture and meets the basic requirements of flexibility and scalability. The model first uses intermediate features from the HBV hydrological physics model, which are used as data for model training to improve the accuracy of predictions. Then, the model adopts an attention-based encoding and decoding LSTM structure to enhance the ability to express the characteristics of hydrological processes, especially in dealing with the sensitivity and accuracy of soil moisture prediction in extreme climate areas, showing better performance than traditional deep learning models. This approach demonstrates the powerful potential of combining deep learning with physical models and provides a new direction for efficient and accurate hydrological models.

[0018] Other features and advantages of the technical solution will be described in the subsequent description, and partly become apparent from the description, or understood by implementing the technical solution. The purpose and other advantages of the technical solution can be achieved and obtained through the structures specifically pointed out in the written description and the drawings.

[0019] The technical solution of the present technical solution is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to provide a further understanding of the technical solution and constitute a part of the specification. Together with the embodiments of the technical solution, they are used to explain the technical solution and do not constitute a limitation of the technical solution. In the accompanying drawings: Figure 1 A diagram of the actual operation process of the AEDLSTM (HBV) model provided by the present invention; Figure 2 R in the present invention 2 , KGE and RMSE(m 3 / m 3 )Box plot; Figure 3 R in the present invention 2 , KGE and RMSE(m 3 / m 3 ) Scatter plot; Figure 4 R in the present invention 2 , KGE and RMSE(m 3 / m 3 )CDF plot; Figure 5 R in the present invention 2 , KGE and RMSE(m 3 / m 3 )Global distribution map; Figure 6 R in the present invention 2 , KGE and RMSE(m 3 / m 3 ) Global distribution improvement map. DETAILED DESCRIPTION

[0021] The preferred embodiments of the present technical solution are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present technical solution, and are not used to limit the present technical solution.

[0022] The present invention provides a soil moisture prediction method based on a physical process attention encoding and decoding LSTM model, comprising the following steps: S100. Obtain the surface soil moisture value and its influencing factors at each historical moment, wherein the influencing factors include atmospheric forcing variables, land surface variables and static variables.

[0023] Specifically, in step S100, the atmospheric forcing variables include two-meter temperature, 10-meter east-west component wind speed, 10-meter north-south component wind speed, precipitation, ground air pressure, and specific humidity; the land surface variables include volume soil water layer, ground solar radiation, ground thermal radiation, soil temperature, and evaporation; and the static variables include soil moisture content, clay, sand, silt, and altitude data DEM.

[0024] The present invention uses the LandBench dataset as the dataset for model training, which provides a standardized platform for predicting the performance of land surface variables (LSVs). Its land surface variables come from the ERA5-land reanalysis dataset, atmospheric forcing variables come from the ERA5 reanalysis dataset, and static variables come from the SoilGrids dataset as well as the soil water content extracted from the literature and the vegetation cover type extracted from the literature.

[0025] Table 1 lists all the data used in the present invention. The surface soil moisture is negligible in volume, equivalent to a water layer with an average thickness of 8 mm covering less than 0.001% of the global land surface, but it is crucial to the development and persistence of extreme weather events (such as droughts, floods and heat waves). Therefore, the predicted variable is the surface soil moisture (0-7cm). The simulation results of the land surface model (LSM) are driven and affected by atmospheric data, and its accuracy and reliability depend on the quality and accuracy of atmospheric forcing data. The atmospheric variables finally selected include two-meter temperature, 10-meter east-west component wind speed, 10-meter north-south component wind speed, precipitation, ground pressure, and specific humidity. The thermal properties of the soil determine the speed and stability of its temperature change, while soil moisture regulates this effect. Surface radiation affects the energy input and output of the soil. At the same time, soil moisture is closely related to evapotranspiration. These variables further regulate energy balance and climate systems. Therefore, the land surface variables use volume soil water layer, ground solar radiation, ground thermal radiation, soil temperature and evaporation. Static variables represent unchanging soil and vegetation information, which directly or indirectly affect the distribution and change of soil moisture. The static variables we used were soil moisture content, clay, sand, silt, and elevation data (DEM) extracted from Yamazaki et al. (2017).

[0026] Table 1. Summary of data sources and variables of the LandBench1.0 dataset models used for evaluation and comparison

[0027] S200, input some influencing factors into the HBV hydrological physical model and run it to obtain intermediate variable data.

[0028] The Hydrologiska Byråns Vattenbalansavdelning (HBV) model is a hydrological model known for its simplicity, high performance, and persistence. The HBV model began as a basic centralized approach and later developed into a distributed model. It includes three main physical components: snow accumulation and melt, soil moisture accounting, and runoff response, including groundwater dynamics. Its success is attributed to three key factors: simplicity, which makes the model easy to understand and reprogram; strong performance, especially evident in comparative studies; and persistence, which has continued to improve over many years of continuous development and application. This simplicity and adaptability make the HBV model applicable to a wide range of climates and terrains, providing a reliable tool for global hydrological modeling. While the core structure of the model remains consistent, many variants have been developed to suit different regions and applications.

[0029] In an achievable manner, the influencing factors input into the HBV hydrological physics model are the two-meter temperature and precipitation in the atmospheric forcing variables and the evaporation data in the land surface variables. The present invention utilizes the above-mentioned two-meter temperature, precipitation and evaporation data to simulate and generate intermediate variable data such as snow melting (melt), refreezing (refreezing), soil moisture content (soil_wetness), recharge (recharge), soil water penetration (tosoil), shallow aquifer (SUZ) and deep aquifer (SLZ) of the HBV model.

[0030] S300: performing correlation analysis between the intermediate variable data and the surface soil moisture, and selecting the intermediate characteristic variable combination most correlated with the surface soil moisture.

[0031] When combining intermediate variables with AEDLSTM, feature selection considerations need to be made. The HBV model is a complex hydrological model. The combination of multiple intermediate variables may contain a lot of information, but it may also introduce noise, resulting in a decrease in model performance. Adding too many features may introduce noise, increase the risk of dimensionality disaster, lead to overfitting and increased computational costs, and reduce model efficiency and generalization capabilities. Therefore, selecting appropriate variables from the intermediate variables of the HBV model to add to deep learning model training is also the key to improving model performance. The greedy nature of the feature selection algorithm may make it impossible to ensure that the selected feature set is optimal, and may select features that are not conducive to model performance. Therefore, the present invention selects correlation analysis and random forests for feature selection to improve the performance and generalization ability of the model.

[0032] By calculating the correlation coefficient, the present invention quantifies the linear correlation between each intermediate variable and soil moisture. Random forest can capture nonlinear relationships and interactions, and measure the contribution of each variable to soil moisture by evaluating its importance in the model.

[0033] The Pearson correlation coefficient is a statistical indicator used to measure the degree of linear correlation between two variables. Its value range is between -1 and 1, indicating the strength and direction of the linear relationship between the two variables. When the correlation coefficient is 1, it indicates a perfect positive correlation; when it is -1, it indicates a perfect negative correlation; when it is 0, it indicates no linear correlation. The calculation formula of the Pearson correlation coefficient is as follows: ; in, and denote the i-th observed value of the two variables, and The Pearson correlation coefficient can help researchers understand and quantify the association between variables by calculating the covariance between two variables divided by their respective standard deviations.

[0034] Random forest is an ensemble learning method that combines multiple independent decision trees to make predictions. During training, random forest uses a "bagging" technique to randomly sample data, and then each decision tree is trained on a different subset of the data, which helps reduce overfitting. The final prediction result is obtained by voting or averaging all decision trees. This ensemble method effectively reduces the variance of the model and improves generalization ability.

[0035] The importance and relevance of the seven intermediate process variables of the HBV model were evaluated using random forest and correlation analysis. As shown in Table 2, soil moisture content (soil_wetness) has the highest feature importance (0.71) and relevance (0.27), while deep aquifer (SLZ) ranks second in feature importance (0.21) and relevance (0.18). These two variables are the most critical factors for predicting soil moisture. Based on these results, soil moisture content and deep aquifer are selected as input variables of the AEDLSTM (HBV) model, aiming to enhance the model's prediction ability through these key physical information. The introduction of these two variables provides the model with important physical process information required for accurate prediction of soil moisture. This step is based on objective data of feature importance and relevance, showing the application value of deep learning models that combine data-driven with physical processes in the field of atmospheric science.

[0036] Table 2. Random forest importance and Pearson correlation coefficient of HBV model features feature importance Correlation coefficient Soil moisture content 0.72 0.27 Deep aquifer 0.21 0.18 Make-up water 0.053 0.033 Water penetration into soil 0.0085 0.034 Snow Melt 0.0031 0.033 Heavy condensation 0.0028 0.057 Shallow aquifer 0 0.0043 Based on the results of these feature selection methods, the present invention selects the combination of feature variables that are most relevant to surface soil moisture. This method provides us with a general, flexible and scalable prediction solution, which can flexibly adjust the combination of feature variables according to the actual situation of the object to be predicted, so as to achieve better prediction results. Here, we use the word "flexible" to mean that the model can both integrate additional soil moisture information and learn complex dependencies in the data without assuming a specific functional form. We use the word "scalable" to mean that the model can be efficiently trained on large data sets and improve as additional data is added. These characteristics provide a basis for exploring the contributions of different geophysical data sets, relaxing model assumptions, and incorporating more data into current predictions.

[0037] S400, taking the combination of influencing factors and intermediate characteristic variables as input data, training a preset attention-based encoding and decoding LSTM model so that the model reaches a preset accuracy, and obtaining a trained soil moisture model; Long short-term memory network (LSTM) is a special recurrent neural network (RNN) with important applications in processing time series data. Compared with the traditional RNN structure, LSTM introduces a gating mechanism. Through structures such as forget gate, input gate and output gate, it effectively solves the problems of gradient vanishing and gradient exploding in traditional RNN, enabling it to better capture long-term dependencies, thereby improving the performance and effect of the model.

[0038] The memory unit operates by selectively remembering and forgetting input data, and consists of an input gate, an output gate, and a forget gate. As it passes through the memory cells, it selectively retains and discards information. The first step in LSTM network processing is to pass through the forget gate , the forget gate controls whether the information in the previous memory unit should be retained or forgotten. Its output value range is between 0 and 1, where 1 means completely retained and 0 means completely forgotten. The calculation of the forget gate is formula (1). Input gate Determines what new information will be stored in the memory cell. It contains a sigmoid function and a tanh function to generate update candidate values. It contains a sigmoid function and a tanh function to generate update candidate values. Input gate The sigmoid function in determines the value to be updated, and the tanh layer generates a potential update vector ; and Calculate according to formula (2) and (3) respectively. is a vector with a value range of (0, 1); , and is a set of learnable parameters defined for the input gate, , and is another set of learnable parameters. Update memory cells It determines how the value in the memory cell is updated. It forgets the past information by multiplying the old memory cell value by the output of the forget gate, and updates the information in the memory cell by multiplying the new candidate value by the output of the input gate. The calculation for updating the memory cell is formula (4). Where ⊙ represents element-by-element multiplication, ⊙ Defines the storage The information in will be forgotten. ⊙ Defines the state that will be added to the unit The new information in the final output gate Determines how the information in the memory cell is passed to the hidden state of the next time step Output and hidden state As shown in equations (5) and (6). is a vector with a value range of (0, 1). , and are three learnable parameters defined for the input gate.

[0039] ;

[0040] In the present invention, since soil moisture data exhibits time dependence and complex nonlinear patterns, and is affected by factors such as season, climate, and terrain, it is necessary to effectively capture hysteresis effects and nonlinear relationships. LSTM is selected as a neural network algorithm for processing time series data such as soil moisture observations. Many modern machine learning software packages provide LSTM layers as standard functions, such as PyTorch. PyTorch is an open source Python machine learning library originally developed by the artificial intelligence research team of Meta Platforms and is now affiliated with the Linux Foundation. It is widely used in artificial intelligence fields such as computer vision and natural language processing. Therefore, the present invention uses the LSTM layer component built into the PyTorch framework.

[0041] In order to further improve the accuracy of soil moisture prediction, especially in challenging predictions in extreme climate regions, the present invention takes into account that encoder-decoder LSTM and attention mechanism have been widely verified as effective tools in soil moisture prediction. At the same time, the present invention realizes that the physical model contains rich hydrological information, especially intermediate process variables. In view of this, the present invention proposes a new surface soil moisture prediction model, named AED-LSTM (HBV). The model combines the LandBench dataset with the intermediate variable data output by the HBV hydrological physical model after feature selection as the input of the AEDLSTM model, in the hope of further improving the prediction performance.

[0042] In one achievable manner, the above step S400 specifically includes: S410: Divide the combined data of influencing factors and intermediate characteristic variables into a training set, a validation set, and a test set; S420: Performing model training on a preset attention-based encoding and decoding LSTM model using a training set; S430: verifying the trained soil moisture model through a validation set; S440: Testing the verified soil moisture model through a test set to ensure that the soil moisture model reaches a preset accuracy.

[0043] The training process of the attention-based encoding and decoding LSTM model in step S420 specifically includes: S421: The combination of influencing factors and intermediate feature variables is used as input data and processed by the encoder LSTM layer, so that the encoder effectively extracts the spatiotemporal feature representation in the input data and converts it into a hidden state of fixed dimension; S422: Use a multi-head attention layer to weight the output of the encoder, highlight key features, and establish effective interactive relationships between features; S423: Perform linear transformation on the attention-weighted output data through the fully connected layer to further integrate and transform the feature representation; S424: Process the output data through the decoder LSTM layer, use the encoder's hidden state information and previous prediction results to gradually generate an output sequence, and randomly discard the output data; S425: Perform a linear transformation on the output of the last time step through a fully connected layer to obtain the final decoder output result.

[0044] like Figure 1 As shown, this is a process diagram of the actual operation of the above-mentioned AEDLSTM (HBV) of the present invention. The present invention uses the temperature, precipitation and evapotranspiration data from 2000 to 2020 in the data set as the input of the HBV model, and runs the HBV model to obtain 20 years of intermediate variable data, and uses these variable data to calculate the correlation coefficient and random forest importance between them and surface soil moisture. Therefore, we select the optimal combination of these variables according to the correlation and importance ranking for experimentation. Finally, the land surface variables, atmospheric forcing variables, static variables and the optimal intermediate variables are used as input data for training, verification and testing.

[0045] First, the input data is processed by the encoder LSTM layer, and the encoder can effectively extract the spatiotemporal feature representation in the input data and convert it into a hidden state of fixed dimension. Then, the encoder output is weighted by the multi-head attention layer to highlight the key features and establish an effective interaction relationship between the features. Subsequently, the attention weighted output is linearly transformed by the fully connected layer to further integrate and transform the feature representation. Then, the output data is processed by the decoder LSTM layer, and the output sequence is gradually generated using the encoder's hidden state information and previous prediction results. After that, in order to prevent overfitting problems, the output data is randomly discarded. Finally, the output of the last time step is linearly transformed by the fully connected layer to obtain the final decoder output result, completing the entire prediction process. Such a model design and processing flow effectively combines the encoder's feature extraction ability, the key feature weighting and interaction of the attention mechanism, the decoder's sequence decoding ability, and the prevention measures for overfitting problems, thereby improving the performance and generalization ability of the model in the soil moisture prediction task.

[0046] The encoder-decoder structure is better than the traditional LSTM model in tasks such as machine translation. Its advantages are mainly reflected in the flexibility and better information management capabilities when processing input and output sequences of different lengths of hydrological data. It consists of two key parts: the encoder and the decoder. The encoder converts the entire input sequence into an intermediate context vector, so that the model can better capture global information and effectively deal with the complex relationship between input and output, while the decoder decodes this vector into a target sequence. In addition, the encoder-decoder structure can also better handle long sequences, making the prediction results more accurate while reducing the burden of long-term dependencies, which makes it more advantageous in the prediction of hydrological data. Considering the successful application of the encoder-decoder structure in other fields of hydrology, such as excellent performance in tasks such as runoff prediction, the encoder-decoder structure is also expected to show better performance in the soil moisture prediction model that uses the intermediate variables in the HBV physical model as additional outputs.

[0047] The attention mechanism (AM) is a solution for arranging resources in the field of computer vision, selecting key information to focus on to obtain the best possible output. It is also well-known in sequence-to-sequence tasks. For each output time step, all input time steps on the encoder side are weighted summed to focus on the part of the input sequence that is relevant to the current output. Here, the present invention uses a multi-head attention layer, an extension of the attention mechanism. Through multi-head attention, the model can focus on different features from many representations of multi-head attention. In hydrological forecasting, this means that the model can more effectively learn the relationship between each time step in the input sequence, such as seasonal changes, periodic changes, etc. The formula of the multi-head attention mechanism is as follows: ; ; ;

[0048] The attention score calculated by each head is calculated by multiplying the query vector Q, key vector K, and value vector V by the corresponding weight matrix respectively, and then summing them up. The attention score is used as the output value of the corresponding position after passing through the softmax activation function. The dimension of key K is 128, It is The weight matrix of each head, is the output weight matrix. The query Q, key K, and value V are all the same, that is, they are all the output of the encoder. This operation is usually called self-attention. This setting is very helpful in capturing the importance of each position in the sequence, which helps the model better understand the dependencies and important information in the sequence. In the field of hydrological forecasting, the attention mechanism is widely used because of its ability to effectively handle complex hydrological data features.

[0049] In one achievable method, the present invention uses 1° spatial resolution data from 2000 to 2020. It is found that the resolution of the data set has no effect on the trend of model performance improvement; the 1° resolution avoids processing too much data and meets the needs of model training. The above data set is divided into three parts: the data from 2000 to 2019 are used for training sets and validation sets, and the data from 2020 are used as test sets. The ratio of training sets to validation sets is 4:1. In order to improve computational efficiency, all training data are collected by random sampling of all grids. The original DEM resolution is 3″ (90m), and the Köppen-Geiger climate classification map uses a 0.5° resolution. In order to match the LandBench data, bilinear interpolation is used in this embodiment to interpolate all variables to 1° resolution, covering a total of 180×360 grid points. In order to speed up the convergence of the model, the minimum and maximum normalization method is used to scale the data to the range of [0, 1]. The specific formula is as follows: ; Where X, are the original value, maximum value, minimum value and normalized value of the training data on the grid respectively.

[0050] Through a large number of experimental processes, the present invention finally determined a set of optimized long short-term memory (LSTM) model parameters. Specifically, the learning rate is set to 0.001 to promote the stable convergence of the model during training; the hidden unit size is 128 and the batch size is set to 64, which effectively reduces the risk of overfitting; the training cycle is set to 1000 to ensure that the model can deeply learn the complex characteristics of the data set; the number of iterations is 400, which ensures sufficient parameter updates in each training cycle and helps to find the best weight configuration; the discard rate is set to 0.15 to prevent the model from being overly sensitive to any single feature or pattern in the training data; the head of the multi-head attention mechanism is set to 2, which can maintain the generalization ability of the model. Finally, the sequence length is 365 days, which enables the model to recognize seasonal changes and long-term dependencies throughout the year. All experiments were conducted on a server equipped with an Intel Core (TM) i9-10980XE, 3.00 GHz × 36 CPU, 128 GB memory, and two NVIDIA RTX A800 graphics cards.

[0051] During all model training and testing processes, the present invention selects and fixes random seeds multiple times. Fixed seeds reduce random differences caused by initial parameters to fairly and accurately compare the performance of different models. At the same time, this strategy ensures consistency of results, enhances experimental reproducibility, and facilitates other researchers to verify and expand the work of the present invention. This rigorous experimental design greatly improves the credibility and repeatability of the research, and effectively promotes the progress of scientific research.

[0052] S500, predicting the surface soil moisture at a future time by using the trained soil moisture model.

[0053] Performance evaluation of the prediction model: The present invention selects the coefficient of determination (R 2 ), Kling-Gupta efficiency (KGE) and root mean square error (RMSE) are used as performance evaluation indicators. 2 ) is used to verify the percentage of variance explained by the forecast model, Kling-Gupta efficiency (KGE) is used to evaluate the goodness of fit based on correlation, conditional bias, and systematic bias, and root mean square error (RMSE) is used to observe the volatility of the forecast. The formula is as follows: ; ; ; and Respectively represent The actual and predicted values ​​for each time step. and are the averages of the actual and predicted values, respectively. is the linear correlation coefficient between the actual value and the predicted value, is the ratio coefficient between the actual value and the predicted value, is the time series correlation coefficient between the actual value and the predicted value.

[0054] In the field of hydrological forecasting, the long short-term memory network (LSTM) and its enhanced version, the LSTM with an encoder-decoder structure (EDLSTM), are both very influential, so I chose them as the benchmark for evaluating our advanced model AEDLSTM (HBV). In order to make a detailed comparison, this paper not only analyzes the traditional LSTM and EDLSTM models, but also specifically considers the versions that combine the physical hydrological model (HBV) as an intermediate variable, namely LSTM (HBV) and EDLSTM (HBV). In addition, in order to more comprehensively evaluate the impact, the present invention also tests the AEDLSTM model without integrating the HBV variable. Through the performance comparison of these models, we can deeply explore the role of the HBV variable in enhancing the accuracy of soil moisture prediction. At the same time, this also helps us evaluate the potential advantages of AEDLSTM in processing HBV variables over traditional LSTM and EDLSTM models, thereby providing scientific basis and technical support for future model development and application.

[0055] Comparative analysis of the utilization efficiency of HBV intermediate variables and AED mechanisms in various models Table 3. R², KGE and RMSE (m 3 / m 3 ) Median value of the indicator Model <![CDATA[R 2 ]]> KGE RMSE AEDLSTM(HBV) 0.892 0.888 0.0161 AEDLSTM 0.833 0.844 0.0195 EDLSTM(HBV) 0.860 0.876 0.0182 EDLSTM 0.857 0.866 0.0184 LSTM(HBV) 0.834 0.836 0.0197 LSTM 0.831 0.843 0.0199 It can be clearly seen from Table 3 that when the HBV intermediate variable is not added, the EDLSTM model is better than the original LSTM model in R 2 and KGE indicators are slightly improved (about 3%), and RMSE indicators are significantly improved (about 7.5%). 2 The KGE index slightly decreased (about 2.5%-2.8%), and the RMSE index degraded (about 6%). Adding the encoding-decoding structure has a positive impact on model performance, but adding the attention mechanism may not always be able to further improve model performance, and sometimes may even lead to performance degradation. It needs to be evaluated and adjusted according to the specific situation.

[0056] After adding the HBV intermediate variable, AEDLSTM, EDLSTM, and LSTM all improved in all three indicators. This improvement is particularly prominent in the AEDLSTM (HBV) model, with R 2The value increased from 0.8328 to 0.8922 (improved by about 7.13%), the KGE value increased from 0.8440 to 0.8882 (improved by about 5.24%), and the RMSE value also improved significantly, from 0.0195 to 0.0161 (improved by about 17.44%). In contrast, although the EDLSTM and LSTM models also improved after adding HBV variables, the improvement was much smaller than that of AEDLSTM. This shows that the AEDLSTM model makes more effective use of HBV intermediate variables and can better capture and integrate this information to improve the accuracy of prediction. In addition, it is worth noting that when the HBV intermediate variables are not integrated, the performance of AEDLSTM is lower than that of the EDLSTM model. This may suggest that without the HBV intermediate variables, the attention encoding and decoding mechanism of AEDLSTM fails to fully function or that the model may be too dependent on HBV variables, resulting in poor performance compared to EDLSTM that focuses on the basic LSTM structure. However, when the HBV intermediate variables are introduced, the performance of AEDLSTM is greatly improved, which shows its great potential in utilizing the information of these physical processes. By effectively integrating the HBV intermediate variables, AEDLSTM (HBV) not only improves its own prediction accuracy, but also highlights the powerful advantages of hybrid physical knowledge and deep learning methods, providing strong evidence for the further development and application of hydrological models.

[0057] The box plot provides a robustness evaluation method that does not rely on the assumption of normal distribution of data, helping us identify outliers and making model comparison more objective and accurate. To fully demonstrate the performance of different deep learning models in hydrological forecasting, Figure 2 As shown in the figure, the present invention uses box plots to intuitively show the performance of each model in R 2 , KGE and RMSE, including the distribution range, median, quartiles and outliers of the data. This graphical approach highlights the stability of the model and its sensitivity to outliers. 2As can be seen from the box plots, the variants with the HBV model as input (green, light blue, and blue) generally have more compact boxes, indicating that the model with HBV may provide more consistent prediction performance. The AEDLSTM model (red) has a lower median, but after adding HBV (blue), the median is improved and the boxes are more compact, indicating that the combination of attention mechanism and HBV can effectively improve model performance. The performance of the encoder-decoder structure LSTM (light green and light blue) does not seem to be much different before and after adding HBV, which may be because when the feature richness is not enough to meet the requirements of the encoder-decoder structure network, the model may overfit the noise in the training data instead of learning generalized features. In contrast, the simple LSTM model may be better at learning with newly introduced and valuable features due to its simpler structure, especially in these cases. In the box plot, the "whiskers" represent the range of variation in model performance, which shows the upper and lower limits of the distribution of data points. Specifically for LSTM(HBV) and AEDLSTM(HBV), the lower whiskers (i.e., the straight lines below the box plots) of both are shorter than those of the corresponding LSTM and AEDLSTM without the HBV model, which indicates that the version with the HBV model has improved in the worst performance. 2 The minimum value is improved compared to the basic LSTM model. For AEDLSTM (HBV), its R 2 The minimum value is also higher than that of the AEDLSTM model without the HBV model. This means that after adding the HBV model, the performance of the model in the most unfavorable situation is better than that without the HBV, that is, the stability and robustness of the model are enhanced. This may mean that in the most difficult prediction situation, the intermediate variables provided by the HBV model help the LSTM model better capture the important features in the data. Similarly, from the box plots of KGE and RMSE, it can be seen that in terms of KGE and RMSE performance indicators, after the LSTM model adds the HBV physical model input, such as EDLSTM (HBV) and AEDLSTM (HBV), its performance consistency has improved, which is manifested as a more compact box and a shorter performance lower bound "whiskers". After adding HBV, the performance of EDLSTM and LSTM models is slightly improved, but still lower than that of AEDLSTM (HBV), highlighting the latter's advantage, and the KGE of the LSTM model after adding HBV has even decreased, indicating that its utilization of HBV intermediate variables needs to be improved.

[0058] Unlike box plots, which are mainly used to show the statistical distribution of a variable, scatter plots can effectively show the distribution of data points by revealing the correlation between predicted values ​​and actual observed values, thereby identifying any linear or nonlinear relationships and outliers in the data. By observing whether the points are closely arranged close to the diagonal, we can intuitively judge whether the model's prediction is accurate. In order to intuitively display and evaluate the relationship between the model's prediction accuracy and the observed data, the present invention draws a scatter plot, such as Figure 3 As shown. Figure 3 As can be seen from the figure, the original LSTM model shows a reliable baseline performance in soil moisture prediction. In the low humidity range (e.g., 0 to 0.2), the scatter plot shows that the predicted values ​​are relatively matched with the observed values, indicating that the model can accurately predict low soil moisture. In the high humidity range (e.g., 0.6 to 0.7), it can be seen that the scatter plot begins to deviate from the 1:1 line, and the prediction accuracy of higher humidity values ​​decreases, indicating that the model is too conservative when predicting soil moisture close to saturation and misestimates the actual humidity value. Figure 3 e. In the high humidity range, the scatter points of the predicted values ​​tend to be closer to the regression line, which can also be seen in the EDLSTM and AEDLSTM figures, indicating that the addition of the HBV intermediate variable has a significant improvement in the prediction ability for the high humidity range, but Figure 3 It can also be seen in Figure e that LSTM (HBV) has some outliers in the low humidity range, indicating that LSTM's ability to utilize HBV intermediate variables in the low humidity range is still insufficient. Figure 3 c and Figure 3 d. It can be seen that in the low humidity range, the scatter density of the EDLSTM model with the encoder-decoder structure is improved in the low humidity range, indicating that it is improved over the ordinary LSTM model. The encoder-decoder structure helps the model better capture the complex dependencies in the time series data, and therefore may provide more accurate humidity predictions. Figure 3 a and Figure 3b. AEDLSTM with attention mechanism shows that the scatter points are more compact and concentrated around the ideal fitting line in the low humidity range. This shows that the prediction of AEDLSTM model is not only more accurate than LSTM model, but also better than EDLSTM model with only encoder-decoder structure. The addition of attention mechanism improves the model's ability to identify important features in input data. Especially in low humidity, small changes in data are also important, and attention mechanism can help the model focus on these subtle changes. Therefore, AEDLSTM can provide more accurate low humidity prediction, which is reflected in more compact data distribution and lower prediction error compared with the model without attention mechanism. Due to the combination of HBV intermediate variables and attention mechanism, AEDLSTM (HBV) shows better prediction ability at extreme values ​​of high and low humidity, which is particularly important in hydrological modeling, because prediction errors at these extreme values ​​may lead to inaccurate prediction of hydrological events, which is crucial for responding to natural disasters and protecting ecosystems.

[0059] In order to fully understand and compare the statistical distribution of predictions from different models, as well as to accurately identify trends, skewness, and outliers in the data, we plotted the cumulative distribution function (CDF) graph, as shown in Figure 4 This graph shows the cumulative frequency of data values, providing an intuitive way to observe the percentage of data at any given threshold. We can easily compare the distribution differences of data under different models or experimental conditions, as well as quickly determine the percentile of the data, such as the median or other specific probability thresholds. From the cumulative distribution function (CDF) Figure 4 a It can be seen that the CDF curves of all models are closely clustered, which means that they are in R 2 The performance of the two models is very close, but some differences can still be seen. AEDLSTM(HBV) shows an overall lead, and its CDF curve is closer to the upper right corner, indicating that this model achieves a higher R in most cases. 2 It is followed by EDLSTM(HBV) and EDLSTM, which perform similarly, but have a higher R 2 Value range, from Figure 4 In b, i.e. the enlarged part marked by the red box, the performance of AEDLSTM (HBV) is more concentrated in high R 2 , indicating that it is more consistent with the other models in terms of optimal performance. Figure 4 c and its enlarged part marked by a red box Figure 4 d, Figure 4 e and its enlarged part marked by a red box Figure 4 f, it can also be seen that AEDLSTM (HBV) performs best in KGE and RMSE indicators, and it provides predictions with higher KGE and lower errors in a wider range.

[0060] Through the above series of detailed analyses, it can be concluded that incorporating the intermediate variables of the HBV model into the deep learning model (especially the AEDLSTM structure that integrates the attention mechanism encoder-decoder LSTM) can significantly improve the performance of the model in hydrological prediction. These findings not only verify the positive impact of the integration of HBV intermediate variables, but also emphasize the importance of integrating machine learning and physical process knowledge to improve model accuracy. In particular, in the AEDLSTM (HBV) model, the attention mechanism is integrated into its structure, which greatly enhances the model's prediction ability and demonstrates the great potential and application value of combining data-driven and physical-driven in hydrological model prediction.

[0061] AEDLSTM (HBV) global distribution and improved performance evaluation To comprehensively demonstrate the model's performance in predicting soil moisture at a global scale, we plotted KGE, R 2 These charts not only allow us to intuitively understand the performance patterns of the models in geographic space, but also help to comprehensively evaluate the overall performance of the models. By comparing the distribution maps of different regions, we can identify the accuracy and stability of the models in specific climate zones and terrain conditions, and provide effective guidance for decision support, model improvement and scientific communication. Figure 5 The R values ​​of all models for surface soil moisture prediction are shown in 2 , KGE, and RMSE. To evaluate in detail the performance improvement of the AEDLSTM (HBV) model over the other five models worldwide and to identify which regions benefit the most, we plotted R 2 , KGE and RMSE global distribution improvement diagram, such as Figure 6 These plots provide an intuitive way to quantify improvements and highlight key areas where model performance differs significantly, providing a basis for further model refinement and decision support. Figure 6 The improvement of AEDLSTM (HBV) over the other five models is shown.

[0062] When comparing model performance, this paper focuses on evaluating the performance of the AEDLSTM (HBV) model relative to the AEDLSTM model on a global scale. 2 Improved effect. Improved ratio through comprehensive analysis Figure 6 , we found that the AEDLSTM(HBV) model achieved positive improvements in most regions, especially in North America, Europe, most of Asia, and Australia. The AEDLSTM(HBV) model performed better than the AEDLSTM model in R 2Significant improvements have been made in the indicators, showing a higher degree of prediction fit to the real observations. The wide distribution of red areas demonstrates the strong generalization ability of the AEDLSTM (HBV) model, which can maintain its prediction accuracy even in a global scale with extremely complex climate, terrain and land use. The breadth and density of these improved areas indicate that by integrating HBV intermediate variables, the deep learning model can more accurately capture the dynamic changes of soil moisture, which reflects the importance of physical process knowledge in improving the model's prediction ability. Figure 6 It can be seen that in certain areas, especially high-latitude and high-altitude areas with less human activities, the performance of the AEDLSTM (HBV) model in predicting soil moisture has been particularly improved. These areas include Canada and Alaska in North America, high-latitude areas in northern Europe, the Qinghai-Tibet Plateau in Asia, the Andes Mountains in South America, and the Sahara Desert in central Africa. In these places with harsh environmental conditions, the AEDLSTM (HBV) model has demonstrated its superior prediction ability compared to the AEDLSTM model. This improvement is likely due to the fact that in sparsely populated areas, the intermediate variables of the HBV model can more fully reflect the natural hydrological process because these areas are less affected by human activities. By incorporating these intermediate variables, the AEDLSTM model can capture richer natural hydrological cycle information, thereby improving prediction accuracy. These areas usually have unique hydrological characteristics. For example, high-latitude and high-altitude areas may have complex snow melting processes, while arid areas have their own unique water evaporation and precipitation patterns. The structure of the AEDLSTM(HBV) model enables it to better capture and utilize these features, while traditional models may lack the ability to handle these complex hydrological signals.

[0063] Combined with these findings, we can have a deeper understanding of the important role of HBV intermediate variables in improving model performance. Especially in the development of hydrological models, the combination of physical process knowledge and machine learning technology is regarded as the key to improving model generalization ability and prediction accuracy. By integrating HBV intermediate variables in the model, not only can the model performance be improved globally, but also significant performance advantages can be achieved in areas where natural environmental factors dominate the hydrological process. Looking forward to the future, such methods that combine physical knowledge and deep learning will provide more accurate and reliable prediction tools for the field of hydrological science.

[0064] In summary: Deep learning soil moisture prediction models have been particularly popular recently, but due to the high spatial heterogeneity of soil texture, the prediction ability of deep learning models still needs to be improved in high latitudes and high altitudes, as well as in areas with harsh arid environmental conditions such as deserts. There are a large number of intermediate variable data in traditional physical hydrological models. These nonlinear data also contain a large amount of hydrological information, but there is also noise information. How to use these intermediate variable data may affect the prediction results. The present invention designs a novel AEDLSTM (HBV) model to improve the prediction performance of deep learning models in extreme geographical and geological environments. The proposed model designs the intermediate variable selection of the HBV hydrological model, and uses the encoding-decoding long short-term memory model (AEDLSTM) with attention to most accurately predict soil moisture, which is superior to the LSTM and EDLSTM models in terms of statistical indicators of soil moisture prediction. It is recommended to use the following hyperparameters when predicting soil moisture in the AEDLSTM (HBV) model: epochs of 1000, hidden unit size of 128, batch size of 64, learning rate of 1e-4, and time step of 365.

[0065] The AEDLSTM (HBV) model combines an LSTM encoder, an LSTM decoder, and an attention mechanism to model and generate sequence data. The intermediate process variables of the HBV hydrological model are added to the input data, and this information further enriches the model's input. First, the model encodes these rich input sequences into hidden state sequences through the LSTM encoder layer. Then, through the multi-head attention layer, the model can better understand the importance of each part of the input sequence and pass the attention-weighted information to the decoder. The decoder part uses this information to generate the final output sequence. During the entire training process, in order to reduce the risk of overfitting, the model randomly discards some data through the Dropout layer. In summary, the AEDLSTM model aims to improve the modeling and generation of sequence data by combining multiple information sources. It is particularly suitable for scenarios that need to consider multiple variables and complex associations, such as prediction and simulation tasks in the HBV hydrological model.

[0066] According to the experimental results, from R 2 , KGE and RMSE performance indicators, the AEDLSTM (HBV) model performs best, with its R 2 The value is 0.8922, the KGE value is 0.8882, and the RMSE is 0.0161, showing higher prediction accuracy than other models.

[0067] The results show that the AEDLSTM (HBV) model performs well in soil moisture prediction, especially in extreme cases of low and high humidity, which is better than other models. It combines the HBV intermediate variables and the attention mechanism to more accurately capture the changes during the wet and dry seasons. It performs significantly globally, especially in North America, Europe, Asia and Australia, emphasizing the importance of combining physical process knowledge with deep learning technology, and provides a new method for prediction research in the field of hydrological science. This comprehensive application provides a more reliable and accurate method for soil moisture prediction, which is of great significance to fields such as water resources management and agricultural planning.

[0068] Obviously, those skilled in the art can make various changes and modifications to the technical solution without departing from the spirit and scope of the technical solution. Thus, if these modifications and variations of the technical solution fall within the scope of the claims of the technical solution and their equivalents, the technical solution is also intended to include these modifications and variations.

Claims

1. A soil moisture prediction method based on physical process attention encoding and decoding LSTM model, characterized in that: include: Obtaining the surface soil moisture value and its influencing factors at each historical moment, wherein the influencing factors include atmospheric forcing variables, land surface variables and static variables; Input some of the influencing factors into the HBV hydrological physical model and run it to obtain intermediate variable data; Performing correlation analysis between the intermediate variable data and the surface soil moisture, and selecting the intermediate characteristic variable combination most correlated with the surface soil moisture; The influencing factors and the intermediate characteristic variables are combined as input data, and a preset attention-based encoding and decoding LSTM model is trained so that the model reaches a preset accuracy, thereby obtaining a trained soil moisture model; The surface soil moisture at future times is predicted using the trained soil moisture model.

2. The soil moisture prediction method based on the physical process attention encoding and decoding LSTM model as claimed in claim 1 is characterized in that: The atmospheric forcing variables include two-meter temperature, 10-meter east-west component wind speed, 10-meter north-south component wind speed, precipitation, surface air pressure, and specific humidity; The land surface variables include volume soil water layer, ground solar radiation, ground thermal radiation, soil temperature and evaporation; The static variables include soil moisture content, clay, sand, silt and elevation data from DEM.

3. The soil moisture prediction method based on the physical process attention encoding and decoding LSTM model as claimed in claim 2 is characterized in that: The influencing factors input into the HBV hydrological physics model are the two-meter temperature and precipitation in the atmospheric forcing variables and the evaporation data in the land surface variables.

4. The soil moisture prediction method based on the physical process attention encoding and decoding LSTM model as claimed in claim 3 is characterized in that: The intermediate variable data are snow melting, recondensation, soil moisture content, water recharge, water penetration into the soil, shallow aquifer and deep aquifer data.

5. The soil moisture prediction method based on the physical process attention encoding and decoding LSTM model as claimed in claim 4 is characterized in that: The intermediate characteristic variables are a combination of soil moisture content and deep aquifer data.

6. The soil moisture prediction method based on the physical process attention encoding and decoding LSTM model as claimed in claim 1, characterized in that: The correlation analysis includes Pearson correlation coefficient calculation and random forest importance analysis.

7. The soil moisture prediction method based on the physical process attention encoding and decoding LSTM model as claimed in claim 6 is characterized in that: The calculation formula of the Pearson correlation coefficient is as follows: ; in, and denote the i-th observed value of the two variables, and They represent the average values ​​of two variables respectively, and ∑ represents the summation operation.

8. The soil moisture prediction method based on the physical process attention encoding and decoding LSTM model as claimed in claim 1, characterized in that: The influencing factors and the intermediate feature variable combination are used as input data to train the preset attention-based encoding and decoding LSTM model so that the model reaches a preset accuracy, including: Dividing the influencing factors and the intermediate characteristic variable combination data into a training set, a validation set and a test set; Performing model training on the preset attention-based encoding and decoding LSTM model through the training set; Verifying the trained soil moisture model through the verification set; The verified soil moisture model is tested using the test set to ensure that the soil moisture model reaches a preset accuracy.

9. The soil moisture prediction method based on the physical process attention encoding and decoding LSTM model as claimed in claim 8, characterized in that: The training process of the attention-based encoding and decoding LSTM model specifically includes: The influencing factors and the intermediate feature variable combination are used as input data and processed by the encoder LSTM layer, so that the encoder effectively extracts the spatiotemporal feature representation in the input data and converts it into a hidden state of fixed dimension; Using a multi-head attention layer to weight the output of the encoder, highlight key features, and establish effective interactive relationships between features; The attention-weighted output data is linearly transformed through the fully connected layer to further integrate and transform the feature representation; Processing the output data through the decoder LSTM layer, using the encoder's hidden state information and previous prediction results to gradually generate an output sequence, and randomly discarding the output data; The output of the last time step is linearly transformed through the fully connected layer to obtain the final decoder output result.

10. The soil moisture prediction method based on the physical process attention encoding and decoding LSTM model as claimed in claim 8, characterized in that: The parameter settings of the attention-based encoding and decoding LSTM model are as follows: the learning rate is set to 0.001, the hidden unit size is 128, the batch size is 64, the training cycle is 1000, the number of iterations is 400, the drop rate is 0.15, the number of heads of the multi-head attention mechanism is set to 2, and the sequence length is 365 days.

Citation Information

Patent Citations

  • A method for predicting concrete durability based on data mining and artificial intelligence algorithm

    AU2020101854A4

  • Soil humidity prediction method and device based on EDC-LSTM model and storage medium

    CN114386332A

  • Traffic flow prediction method based on improved space-time Transform

    CN115273464A

  • Lithium battery residual life probability prediction method based on time convolution attention mechanism

    CN116381517A

  • Soil texture prediction method based on local-global dependency relationship

    CN116881664A

Cited By

  • Method and device for calculating spatial non-stationary soil humidity

    CN119827740A

  • Soil organic matter content prediction method based on two-channel model

    CN119989442A

  • AttnConvLSTM soil moisture content intelligent forecasting method fused with CLDAS

    CN120873509A

  • Wheat soil moisture prediction method based on mixed PPLSTM model

    CN121997038A