Soil moisture content prediction method and device based on deep learning

Abstract

The invention discloses a soil moisture content prediction method and device based on deep learning, and the method comprises the steps: carrying out the seasonal trend decomposition of historical soil moisture content time sequence data, and obtaining a trend term, a seasonal term and a residual term; secondly, analyzing a causal relationship between each decomposition item and a meteorological variable in combination with a plurality of causal test methods, and constructing a dynamic causal adjacency matrix to obtain a causal tree structure; for the trend item, the season item and the residual item, designing a causal LSTM sub-model according to the fruit tree structure, inputting a historical decomposition sequence and the screened causal meteorological variables, and carrying out independent prediction; and finally, fusing prediction results of the sub-models through a full connection layer, and outputting a final prediction value of the soil moisture content in a future period. The innovation point of the method is that causal information and time dependence characteristics of decomposition items are fused, and the interpretability of the model and the response capability of the model to long-term trend and short-term mutation are enhanced.

Description

Technical Field

[0001] The present invention relates to a method for predicting soil moisture content, and particularly to a method and device for predicting soil moisture content based on deep learning. Background Art

[0002] Soil moisture content is an important factor affecting agricultural production, drought disaster warning, and water resource management. With the intensification of global climate change, extreme weather events (such as droughts, heavy rains, etc.) occur frequently, greatly affecting the dynamic changes of soil moisture. Therefore, accurate and timely prediction of soil moisture content is crucial for precision agriculture and efficient utilization of water resources. Currently, soil moisture content prediction mainly relies on physical models and statistical methods, but these methods usually face problems such as scarce data, high model complexity, and insufficient response ability to extreme climate conditions.

[0003] Most traditional soil moisture content prediction methods rely on physics-based hydrological models (such as water infiltration and evaporation transport models), which predict changes in soil moisture by simulating the movement of soil water. However, these physical models usually require a large amount of input data and rely on complex soil parameters and environmental factors, resulting in difficult accurate prediction in practical applications, especially under climate change or extreme meteorological conditions. In addition, the computational amount of physical models is large and the processing speed is slow, restricting their application in fast-response scenarios.

[0004] With the development of machine learning technology, especially deep learning methods, soil moisture content prediction has gradually shifted to prediction models based on historical data. As a deep learning model, LSTM (Long Short-Term Memory Network) has been widely used in time series prediction, especially showing its superiority in capturing long-term dependencies. However, the accuracy of the LSTM model is usually restricted by the quality of input data and the modeling ability of the model, especially in the face of complex meteorological driving factors of soil moisture content, its accuracy is greatly reduced. Summary of the Invention

[0005] Object of the Invention: Aiming at the above problems, the present invention proposes a method and device for predicting soil moisture content based on deep learning, enhancing the interpretability of the model and the response ability to long-term trends and short-term mutations.

[0006] Technical Solution: The technical solution adopted by the present invention is a method for predicting soil moisture content based on deep learning, including: Obtaining historical time series of soil moisture content and historical time series of each meteorological factor; According to the historical time series of soil moisture content, using the seasonal trend decomposition method STL to decompose and obtain a decomposed item time series, and the decomposed items include: trend item, seasonal item, and residual item; According to the historical time series of each meteorological factor and the time series of each decomposition item, the prediction results of each decomposition item are obtained through the causal long short-term memory network LSTM sub-model corresponding to each decomposition item. The construction process of the causal long short-term memory network LSTM sub-model includes: according to the historical time series of the trend item and each meteorological factor, through the causal test method, analyzing the causal relationship between the trend item and the meteorological variable, obtaining the dynamic causal adjacency matrix, obtaining the causal tree structure according to the dynamic causal adjacency matrix, and constructing the causal LSTM sub-model according to the causal tree structure. The prediction results of the decomposition items output by each causal LSTM sub-model are fused through the fully connected layer, and the final predicted value of the soil moisture content in the future period is output through the output layer.

[0007] A preferred solution is that the causal long short-term memory network LSTM sub-models corresponding to each decomposition item include: a trend item sub-model, a seasonal item sub-model, and a residual item sub-model, and the architectures of each sub-model are constructed based on the causal LSTM structure. The trend item sub-model adopts a single-item causal LSTM structure. The seasonal item sub-model is provided with a global pooling layer after the seasonal encoder and layer normalization, which is used to extract global seasonal features, and an attention mechanism is introduced before the fully connected layer of the seasonal item sub-model to generate a context vector and attention weights. The residual item sub-model adopts a bidirectional causal LSTM structure, and combines two global feature extraction methods of max pooling and average pooling to obtain the max pooling feature and the average pooling feature, which are connected to the fully connected layer of the residual item sub-model through the tensor connection function after being connected with the residual coding feature.

[0008] A preferred solution is that the causal test method includes the Pearson correlation coefficient, the maximum information coefficient, and the Granger causality test. By calculating the correlation and causal relationship between the time series of the soil moisture content trend item and the meteorological factors, a binary matrix corresponding to each test method is generated according to the set threshold, and the binary matrices are merged by the logical OR strategy to obtain the dynamic causal adjacency matrix. Obtaining the causal tree structure according to the dynamic causal adjacency matrix includes: the causal tree structure expands layer by layer from bottom to top, starting from the leaf node, and connecting the parent nodes that have a causal impact on the current node layer by layer upward; setting the soil moisture content as the leaf node, finding the factors with a value of 1 through the causal adjacency matrix as the parent nodes pointing to the soil moisture content leaf node, and the parent nodes are used as the tree nodes of the second layer; then finding the causal nodes pointing to the tree nodes of the second layer through the causal adjacency matrix as the tree nodes of the third layer until the root node without causal pointing.

[0009] A preferred solution is that the fully connected layer fuses the prediction results of each decomposition item by adaptively adjusting the weights of each decomposition item; the fully connected layer adopts a ReLU activation layer.

[0010] The causal long short-term memory network LSTM sub-model is trained through a data set, and the training process includes: dividing the data set into a training set, a validation set, and a test set. The training set is used for model training, the validation set is used for hyperparameter optimization, and the test set is used to evaluate the prediction effect of the model; the data set includes the time series of each decomposition item and the historical time series of each meteorological factor.

[0011] The historical time series of soil moisture content and the historical time series of each meteorological factor are obtained, and the collection time length is two years or more.

[0012] The present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the soil moisture prediction method based on deep learning described above is implemented.

[0013] The present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the soil moisture prediction method based on deep learning described above is implemented.

[0014] The present invention provides a computer program product, including a computer program and / or instructions. When the computer program and / or instructions are executed by a processor, the soil moisture prediction method based on deep learning described above is implemented.

[0015] The present invention also provides a soil moisture prediction device based on deep learning, including: A data acquisition module, configured to obtain the historical time series of soil moisture content and the historical time series of each meteorological factor; A time series decomposition module, configured to decompose the historical time series of soil moisture content according to the seasonal trend decomposition method STL to obtain the decomposition item time series, and the decomposition items include: a trend item, a seasonal item, and a residual item; A causal analysis module, configured to obtain key meteorological driving factors by analyzing the historical time series of decomposition items and each meteorological factor, including: according to the historical time series of decomposition items and each meteorological factor, through a causal test method, analyzing the causal relationship between each decomposition item and the meteorological variable to obtain a dynamic causal adjacency matrix, and obtaining a causal tree structure according to the dynamic causal adjacency matrix; A causal LSTM prediction module, constructing a causal LSTM sub-model according to the causal tree structure, and configured to obtain the prediction results of each decomposition item through the causal long short-term memory network LSTM sub-model corresponding to each decomposition item according to the historical time series of each meteorological factor and the time series of each decomposition item; A fusion output module, configured to fuse the prediction results of the decomposition items output by each LSTM sub-model through a fully connected layer, and output the final prediction value of the soil moisture content in the future period through an output layer.

[0016] Beneficial effects: Compared with the prior art, the present invention has the following advantages: By STL decomposition, the multi-time scale characteristics of soil moisture data are effectively extracted; a variety of causal monitoring methods are comprehensively used to dynamically screen meteorological driving factors, improving the accuracy and adaptability of causal relationships and enhancing the interpretability of the model; by fusing STL, causal information and time-dependent characteristics, the prediction accuracy of soil moisture is optimized, effectively solving the problem of insufficient prediction accuracy of traditional prediction models under complex meteorological conditions, significantly improving the stability of long-term trends and the response ability to short-term mutations (such as extreme weather events), and providing reliable technical support for precision agricultural irrigation, drought and flood warning, etc. The innovation of the present invention lies in fusing causal information and time-dependent characteristics, enhancing the interpretability of the model and the response ability to long-term trends and short-term mutations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flowchart of the soil moisture prediction method based on deep learning according to the present invention; Figure 2 is a flowchart of STL decomposition; Figure 3 is a schematic diagram of the causal LSTM architecture according to the present invention; Figure 4 is a comparison of the results of soil moisture process prediction by different soil moisture prediction methods. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The technical solutions of the present invention will be further described below with reference to the drawings and embodiments.

[0019] Embodiment 1: The soil moisture prediction method based on deep learning according to the present invention has a flowchart as Figure 1 shown. Taking the data collected at Station A in a certain basin from 2015 to 2023 as an example, the specific implementation steps of the soil moisture prediction calculation based on the fusion of time series decomposition and causal deep learning of the present invention are as follows (the step numbers are for convenience of description when introducing the specific implementation manner and do not represent a specific sequential relationship between the steps): Step 1: Obtain the time series data of soil moisture and meteorological data at this station. The soil moisture data can come from soil moisture observation stations or satellite soil moisture data, such as the SMAP satellite, etc. These data usually include the date and the corresponding soil moisture values. The meteorological data includes variables such as temperature, precipitation, evaporation, etc. The data acquisition sources are station observation data or satellite data, such as the ERA5-Land dataset, and the dataset also contains the corresponding date information.

[0020] Soil moisture data: The data format is a CSV file, including a date column and a measured soil moisture column. The data can be loaded using the pandas library in Python and preprocessed to ensure the correct date format and no outliers in the numerical columns.

[0021] Meteorological data: Also in CSV format, including a date column (date) and multiple meteorological variables (temperature, precipitation, evaporation). Missing meteorological data can be filled by interpolation and aligned with the soil moisture data by date to ensure the integrity of the time series.

[0022] Step 2: Use seasonal trend decomposition (STL) to decompose the time series data of soil moisture. Decompose the time series data of soil moisture at Station A into three components: trend term, seasonal term, and residual term. The STL decomposition process is as Figure 2 shown. Output the decomposition terms through operations such as detrending and deseasonalizing. Combine the decomposition terms with the meteorological factors after date alignment into a dataset, and divide it into a training set, a validation set, and a test set at a ratio of 7:1:2. The training set is used as the input item for model training, the validation set is used for hyperparameter tuning, and the test set is used to evaluate the final actual effect of model prediction.

[0023] Step 3: Conduct causality tests on the trend term, seasonal term, and residual term of the historical soil moisture time series respectively with different meteorological elements to construct a causal tree structure. The following takes the trend term at Station A as an example for illustration: The Pearson correlation coefficient is used to detect the possible linear correlation between the trend term and different meteorological elements, and the threshold is set to 0.5; the maximum information coefficient is used to detect the possible non-linear relationship between the trend term and different meteorological elements, and the threshold is set to 0.5; the Granger causality test is used to detect the possible time-series causality relationship between the trend term and different meteorological elements, and the p-value threshold is set to 0.05; by calculating the correlation and causality relationship between the soil moisture trend term and meteorological factors, according to the threshold setting, each test method will generate a binary matrix (0 or 1), indicating the causality relationship found by the test method. According to the multiple binary matrices obtained from the analysis, a dynamic causal adjacency matrix is merged. "Dynamic" means that different causal adjacency matrices can be generated by adjusting the threshold set by the test method or the merging strategy. For example, when the Pearson correlation coefficient threshold is 0.7 or 0.5, the matrix results may be different; or when the matrix results are merged with the "logical OR" strategy, they are also different. This is at the causal test level. At the input level, when different meteorological factors are selected, different matrices are generated. The process of merging the adjacency matrix adopts the "logical OR" strategy, that is, if any test method believes that there is a causal relationship, then this relationship is retained in the final matrix. This adjacency matrix is used to determine the causal dependence relationship between each feature, and further construct a causal tree structure. The core idea of the causal tree is to represent the causal relationship between time-series features as a directed tree structure, so that the LSTM can process features according to the causal relationship, rather than simply treating all features equally. The causal tree structure expands layer by layer from bottom to top, starting from the leaf nodes, and connecting the parent nodes that have a causal impact on the current node layer by layer upwards.

[0024] Preferably, the soil moisture is selected as the leaf node, and the factors with a value of 1 are found through the causal adjacency matrix as the parent nodes pointing to it, such as evaporation and precipitation, as the tree nodes of the second layer; and so on, then the causal nodes pointing to factors such as evaporation are found through the adjacency matrix until the root node without causal pointing. A simple example is: air temperature → evaporation → soil moisture. The causal tree structure is used to describe the causal state of the units in the LSTM, that is, different weights are assigned to each variable to make its prediction not completely a black box.

[0025] Step 4: According to the trend term, seasonal term, and residual term obtained by STL decomposition, independent causal LSTM sub-models are constructed for each decomposition term. Taking the construction of the soil moisture trend term causal LSTM sub-model as an example, the specific steps are described as follows: As Figure 3 shown, within each node unit, the structure of the standard LSTM is used to process the input features, the hidden state (h_prev) and the cell state (c_prev) of the previous time step, and the information flow is controlled through the input gate (i), forget gate (f), and output gate (o). By calculating the activation values of the input gate, forget gate, and output gate, the cell state (c) and the hidden state (h) are updated.

[0026] The state update of each node unit depends not only on its own input and the state at the previous time step, but also on the states of its child nodes (i.e., the nodes associated with it in the causal tree). This vertical information transmission mechanism updates the state of the current node by considering the state and influence of the parent node (or upstream node) in the causal relationship tree of soil moisture trend terms and meteorological factors.

[0027] Construct a causal LSTM sub-model according to the causal tree structure, and adopt the LSTM model architecture based on the causal relationship structure. In order to incorporate causal information into the LSTM, for each node introduce a new state (referred to as the causal state, denoted as ), and constrain the hidden state of the LSTM (denoted as ) through the parent node in the causal structure (corresponding to the current node ). Thus, the causal relationship of the node (i.e., ) can be used to constrain the hidden state of the node generated by the original LSTM (i.e., ). Since the causal state contains both time dependence and causal information, the causal state of the leaf node is input into the fully connected layer for prediction in a non-hidden state.

[0028] CLSTM can learn the time dependence and causal information of each node in the causal structure through the following formulas. These formulas can be adjusted or improved as needed.

[0029] For each node at time step , its input includes the feature vector , the hidden state and cell state at the previous time step ( and ), and the causal state of the parent node in the upper layer of the causal structure. represents the set composed of all parent node indices of the node .

[0030] First, use the following formula to generate the hidden state and cell state at time step t ( and ). This design process is consistent with the classic LSTM model: Input gate: ; Forget gate: ; Current cell state: ;

[0031] Output gate: ; Current hidden status: ; Secondly, use the parent node through the following steps The causal state of the hidden state Perform combined constraints to obtain nodes The causal state at the current time step. This process consists of three steps: Step 1: For the parent node set Each node (or index) j in the hidden state and nodes The corresponding weights of the causal state are calculated as follows: ; Step 2: The weights generated by the previous step Integration Node Causal state information of all parent nodes: ; Step 3: Calculate the parent node Combine causal information with the current hidden state The weights of the nodes are generated by weighted summation The causal state of: ; ; ; in represents the weight, Represents deviation, is the S-type function, tanh is the hyperbolic tangent function, Represents point-by-point multiplication. Note that the formulas in steps 2 and 3 do not apply to the root node (i.e., a node without a parent node). The causal state of the root node is Equivalent to hidden state , because there is no information from the causal driver to constrain the hidden state. The causal state of the leaf node is input into the fully connected layer (the number of neurons is 1 and the activation function is tanh) to obtain the predicted value.

[0032] The above weight calculation formula is the one used in this embodiment, and various other formulas can be used. The formula can be adjusted or improved as needed.

[0033] Trend items usually manifest as long-term, smooth changes and do not require complex feature extraction. Therefore, a single-item LSTM structure is used to model trend items to capture long-term changes in the data. This sub-model can handle long-term dependencies in time series data and predict long-term trends in soil moisture.

[0034] To better capture the seasonal changes in soil moisture, the seasonal term sub-model designs a global pooling layer to capture the overall seasonal pattern, and introduces an attention mechanism that can automatically identify key time points in seasonal changes, enabling it to perform well in processing periodic patterns.

[0035] In order to capture short-term fluctuations and noise and improve the model's adaptability to abnormal situations, the residual term sub-model adopts a bidirectional LSTM structure and combines two global feature extraction methods, maximum pooling and average pooling, so that the model can simultaneously capture prominent abnormal fluctuations (through maximum pooling) and overall fluctuation trends (through average pooling).

[0036] Step 5: Fusion the prediction results of the three sub-models to obtain the final soil moisture prediction result. The specific operations are as follows: The fusion process uses a fully connected layer (such as a ReLU activated layer) to process the output of each sub-model to produce a comprehensive prediction result. The aggregator does not simply add the prediction values of the three components, but intelligently combines them through learnable weights, which can adaptively adjust the contribution of each component in different situations.

[0037] During the fusion process, the outputs of each sub-model are first concatenated (that is, the prediction results of the trend term, seasonal term, and residual term are concatenated into a vector), and then processed through multiple fully connected layers. These fully connected layers further improve the prediction accuracy by learning the relationship between the sub-models. The activation function uses ReLU, which can effectively capture nonlinear relationships and improve the expressiveness of the model.

[0038] After being processed by the fusion network, the soil moisture prediction result is finally generated through a linear output layer. The function of this output layer is to map the fused features to the target variable, that is, the numerical prediction of soil moisture. Figure 4 As shown in the figure, the results of soil moisture process prediction using different methods are compared. SCL is the method proposed by this invention, namely STL+Cause and Effect+LSTM; SLSTM is STL+LSTM; CLSTM is Cause and Effect+LSTM, and LSTM is a prediction model that uses long short-term memory network alone. Figure 4 It can be seen that the soil moisture prediction method proposed in the present invention has extremely high accuracy during the long-term smooth decline of soil moisture, and its response ability to short-term mutations is also improved to a certain extent.

[0039] Step 6: Select four indicators, namely R² (coefficient of determination), RMSE (root mean square error), MAE (mean absolute error), and KGE (Kling-Gupta efficiency coefficient), as the accuracy evaluation indicators for the prediction results of this model. The specific formulas are as follows: ; ; ; ; Table 1 Comparison of index results between the SCL of the method of the present invention and other different methods (ablation experiment)

[0040] The above prediction results are multi-step prediction results for the next seven days. In terms of accuracy, single-step prediction can achieve a higher level relatively speaking.

[0041] Example 2: In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the above-mentioned soil moisture prediction method based on deep learning is implemented.

[0042] Example 3: In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned soil moisture prediction method based on deep learning is implemented.

[0043] Example 4: In one embodiment, a computer program product is provided, including a computer program / instructions. When the computer program / instructions are executed by a processor, the above-mentioned soil moisture prediction method based on deep learning is implemented.

[0044] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0045] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the specified functions in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the specified functions in one block or multiple blocks.

[0046] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the specified functions in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the specified functions in one block or multiple blocks.

[0047] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the specified functions in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the specified functions in one block or multiple blocks.

[0048] Embodiment 5: The soil moisture prediction device based on deep learning according to the present invention includes: A data acquisition module: used to obtain soil moisture and meteorological factor data in real time as model inputs.

[0049] A time series decomposition module: executes to extract three components, namely the trend term, seasonal term, and residual term of the historical soil moisture time series through the seasonal trend decomposition method (STL method) as input items for the next module, and the results can also be selectively published to the user terminal.

[0050] A causal analysis module: analyzes the causal relationship between soil moisture and meteorological factors according to the generated causal adjacency matrix, further constructs a causal tree to provide a basis for subsequent causal LSTM modeling, and publishes the causal tree construction results to the user terminal for the user to make causal decision references.

[0051] Causal LSTM Prediction Module: It includes three independent sub-models, which separately process the prediction of trend items, seasonal items, and residual items. Each sub-model uses causal LSTM node units for time series modeling, constructs a causal LSTM model according to the causal tree obtained by the causal analysis module, and transmits causal information. The data set is divided into a training set, a validation set, and a test set in a ratio of 7:1:2. The training set is used as the input item for model training, the validation set is used for hyperparameter optimization, and the test set is used to evaluate the final actual effect of model prediction.

[0052] Fusion Output Module: It fuses the outputs of the sub-models through a fully connected network (such as a ReLU activation layer), calculates the final soil moisture prediction value, and publishes it to the user terminal.

[0053] The detailed execution process of each module can be seen in the specific content of Embodiment 1.

Claims

1. A soil moisture prediction method based on deep learning, characterized in that Including: Obtain the historical time series of soil moisture and the historical time series of each meteorological factor; According to the historical time series of soil moisture, decompose it using the Seasonal-Trend Decomposition method STL to obtain the time series of decomposition terms, where the decomposition terms include: trend term, seasonal term, and residual term; According to the historical time series of each meteorological factor and the time series of each decomposition term, respectively obtain the prediction results of each decomposition term through the causal long short-term memory network LSTM sub-model corresponding to each decomposition term; the construction process of the causal long short-term memory network LSTM sub-model includes: according to the historical time series of the decomposition term and each meteorological factor, through the causal test method, analyze the causal relationship between each decomposition term and the meteorological variable to obtain the dynamic causal adjacency matrix, obtain the causal tree structure according to the dynamic causal adjacency matrix, and construct the causal LSTM sub-model according to the causal tree structure; The prediction results of the decomposition terms output by each causal LSTM sub-model are fused through a fully connected layer, and the final predicted value of the soil moisture in the future period is output through the output layer.

2. The soil moisture prediction method based on deep learning according to claim 1, wherein: The causal long short-term memory network LSTM sub-models corresponding to each decomposition term include: a trend term sub-model, a seasonal term sub-model, and a residual term sub-model, and the architecture of each sub-model is constructed based on the causal LSTM structure; The trend term sub-model adopts a unidirectional LSTM structure; The seasonal term sub-model is provided with a global pooling layer after the seasonal encoder and layer normalization, which is used to extract global seasonal features, and an attention mechanism is introduced before the fully connected layer of the seasonal term sub-model to generate a context vector and attention weights; The residual term sub-model adopts a bidirectional LSTM structure, and combines two global feature extraction methods of max pooling and average pooling to obtain the max pooling feature and the average pooling feature, which are connected to the fully connected layer of the residual term sub-model through a tensor connection function after being connected with the residual coding feature.

3. The soil moisture prediction method based on deep learning according to claim 1, characterized in that: The causal test method includes Pearson correlation coefficient, maximum information coefficient, and Granger causality test; by calculating the correlation and causal relationship between the time series of each decomposition term of soil moisture and the meteorological factor, a binary matrix corresponding to each test method is generated according to the set threshold, and the binary matrices are merged using a logical or strategy to obtain the dynamic causal adjacency matrix; Obtaining the causal tree structure according to the dynamic causal adjacency matrix includes: the causal tree structure expands layer by layer from bottom to top, starting from the leaf node, and connecting the parent nodes that have a causal impact on the current node layer by layer upwards; Set the soil moisture as the leaf node, and find the factor with a value of 1 through the dynamic causal adjacency matrix as the second-layer tree node pointing to the soil moisture leaf node; then find the causal node pointing to the second-layer tree node through the dynamic causal adjacency matrix as the third-layer tree node until the root node without causal pointing.

4. The soil moisture prediction method based on deep learning according to claim 1, wherein: The fully connected layer adaptively adjusts the weights of each decomposition term to fuse the prediction results of each decomposition term; the fully connected layer adopts a ReLU activation layer.

5. The method for predicting soil moisture based on deep learning according to claim 1, characterized in that: The causal long short-term memory network LSTM sub-model is trained through a data set. The training process includes: dividing the data set into a training set, a validation set, and a test set. The training set is used for model training, the validation set is used for hyperparameter optimization, and the test set is used to evaluate the prediction effect of the model. The data set includes the time series of each decomposition item and the historical time series of each meteorological factor.

6. The method for predicting soil moisture based on deep learning according to claim 1, characterized in that: The acquisition of the historical time series of soil moisture content and the historical time series of each meteorological factor is carried out for a time length of two years or more.

7. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the deep learning-based soil moisture prediction method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the deep learning-based soil moisture prediction method according to any one of claims 1 to 6.

9. A computer program product, comprising a computer program and / or instructions, characterized in that, When the computer program and / or instruction is executed by the processor, it implements the deep learning-based soil moisture prediction method according to any one of claims 1 to 6.

10. A soil moisture prediction device based on deep learning, characterized in that, Including: A data acquisition module for acquiring the historical time series of soil moisture content and the historical time series of each meteorological factor; A time series decomposition module for decomposing the historical time series of soil moisture content according to the seasonal trend decomposition method STL to obtain the time series of decomposition items, where the decomposition items include: a trend item, a seasonal item, and a residual item; A causal analysis module for obtaining key meteorological driving factors by analyzing the historical time series of decomposition items and each meteorological factor, including: analyzing the causal relationship between each decomposition item and meteorological variables according to the historical time series of decomposition items and each meteorological factor through a causal test method to obtain a dynamic causal adjacency matrix, and obtaining a causal tree structure according to the dynamic causal adjacency matrix; A causal LSTM prediction module for constructing a causal LSTM sub-model according to the causal tree structure, and for obtaining the prediction results of each decomposition item through the causal long short-term memory network LSTM sub-model corresponding to each decomposition item according to the historical time series of each meteorological factor and the time series of each decomposition item; A fusion output module for fusing the prediction results of the decomposition items output by each causal LSTM sub-model through a fully connected layer and outputting the final prediction value of the soil moisture content in the future period through an output layer.

Images