Soil humidity prediction method and device, medium and product

Abstract

The invention discloses a soil humidity prediction method and device, a medium and a product, and relates to the technical field of humidity prediction, and the method comprises the steps: obtaining target soil humidity time sequence data; the target soil humidity time sequence data comprises true values of soil humidity at all positions of the to-be-predicted area at all historical moments in a historical time period; inputting the target soil humidity time sequence data into a soil humidity prediction model to obtain a predicted value of soil humidity at each position of the to-be-predicted region at each to-be-predicted moment in the to-be-predicted time period; the soil humidity prediction model is obtained by using a training set and training a Uform deep learning network, and the Uform deep learning network is constructed based on a LeWin Transform module. The method improves the prediction precision of the soil humidity.

Description

Technical Field

[0001] The present application relates to the technical field of humidity prediction, and particularly to a method, device, medium and product for predicting soil humidity. Background Art

[0002] Soil humidity is an important part of the surface water cycle and has a crucial impact on agricultural production, ecological environment, and hydrological processes. In the monitoring and early warning of drought meteorological disasters, accurate soil humidity prediction can effectively help decision-makers evaluate the severity of drought and formulate countermeasures. However, soil humidity data has high spatio-temporal variability, and traditional numerical models have significant limitations in describing it, especially in terms of high spatial and temporal resolutions, making it difficult to meet the needs of modern precision agriculture and environmental monitoring.

[0003] In current meteorological operations, the prediction of soil humidity often relies on the soil module in meteorological models, but these models mainly focus on large-scale meteorological processes and fail to fully consider the dynamic changes of soil humidity at small scales. Especially in the early stage of drought occurrence, the changes in soil humidity respond rapidly to changes in the external environment, but these subtle changes are easily ignored by traditional prediction methods. In addition, the spatial sparsity of observation stations further increases the difficulty of accurate prediction, especially in areas with complex terrain or scarce observation resources, where the spatio-temporal characteristics cannot be fully considered, resulting in low prediction accuracy for soil humidity. Summary of the Invention

[0004] The purpose of the present application is to provide a method, device, medium and product for predicting soil humidity to solve the problem of low prediction accuracy of soil humidity.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In the first aspect, the present application provides a method for predicting soil humidity, including:

[0007] Obtain target soil humidity time series data; the target soil humidity time series data includes: the true values of the soil humidity at each position in the area to be predicted at each historical moment in the historical period;

[0008] Input the target soil humidity time series data into a soil humidity prediction model to obtain predicted values of the soil humidity at each position in the area to be predicted at each predicted moment in the predicted period; the soil humidity prediction model is obtained by training a training set and a Uformer deep learning network, and the Uformer deep learning network is constructed based on the LeWinTransformer module.

[0009] Optionally, the training process of the soil humidity prediction model includes:

[0010] Obtain the training set; the training set includes multiple groups of training data, and the training data includes: the true values of the soil moisture at each position of the sample area at each historical moment in two adjacent historical periods;

[0011] Construct the Uformer deep learning network based on the LeWin Transformer module;

[0012] Use the training set to train the Uformer deep learning network to obtain the soil moisture prediction model.

[0013] Optionally, the Uformer deep learning network includes: an input projection module, an encoder, a bottleneck layer, a decoder, and an output projection module connected in sequence, and the encoder and the decoder are skip-connected.

[0014] Optionally, the encoder includes multiple LeWin Transformer modules and multiple downsampling modules, the bottleneck layer includes 1 LeWin Transformer module, the decoder includes multiple LeWin Transformer modules and multiple upsampling modules, and the number of LeWin Transformer modules and downsampling modules in the encoder is equal to the number of LeWin Transformer modules and upsampling modules in the decoder.

[0015] Optionally, inputting the target soil moisture time series data into the soil moisture prediction model to obtain the predicted values of the soil moisture at each position of the area to be predicted at each moment to be predicted in the period to be predicted, includes:

[0016] Input the target soil moisture time series data into the input projection module in the soil moisture prediction model for convolution operation to obtain the initial features of the area to be predicted;

[0017] Input the initial features of the area to be predicted into the encoder in the soil moisture prediction model for multiple windowed self-attention operations, local enhancement, and downsampling processing to obtain the encoded features of the area to be predicted;

[0018] Input the encoded features of the area to be predicted into the bottleneck layer in the soil moisture prediction model for windowed self-attention operation and local enhancement processing to obtain the features of the area to be predicted after being processed by the bottleneck layer;

[0019] Input the features of the area to be predicted after being processed by the bottleneck layer into the decoder in the soil moisture prediction model for multiple windowed self-attention operations, local enhancement, and upsampling processing to obtain the decoded features of the area to be predicted;

[0020] Input the decoded features of the area to be predicted into the output projection module in the soil moisture prediction model for convolution operation to obtain the predicted values of the soil moisture at each position in the area to be predicted at each moment to be predicted during the period to be predicted.

[0021] Optionally, training the Uformer deep learning network using the training set to obtain the soil moisture prediction model includes:

[0022] Based on the power spectral density and the mean square error, using the true values of the soil moisture at each position in the sample area at the previous historical moment in each group of training data in the training set as the input, and using the true values of the soil moisture at each position in the sample area at the corresponding next historical moment as the output, training the Uformer deep learning network to obtain the soil moisture prediction model.

[0023] Optionally, after training the Uformer deep learning network using the training set to obtain the soil moisture prediction model, it further includes:

[0024] Evaluating the performance of the soil moisture prediction model using the mean square error.

[0025] In a second aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the soil moisture prediction method described in any one of the above.

[0026] In a third aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the soil moisture prediction method described in any one of the above.

[0027] In a fourth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the soil moisture prediction method described in any one of the above.

[0028] According to the specific embodiments provided by the present application, the following technical effects are disclosed in the present application:

[0029] The present application discloses a soil moisture prediction method, device, medium and product. First, target soil moisture time series data is obtained; the target soil moisture time series data includes: the true values of the soil moisture at each position in the area to be predicted at each historical moment in the historical period. Then, the target soil moisture time series data is input into the soil moisture prediction model to obtain the predicted values of the soil moisture at each position in the area to be predicted at each predicted moment in the predicted period; the soil moisture prediction model is obtained by training the Uformer deep learning network using a training set, and the Uformer deep learning network is constructed based on the LeWinTransformer module. The present application uses the soil moisture prediction model obtained based on the LeWin Transformer module to predict the soil moisture, improving the prediction accuracy of the soil moisture. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0031] Figure 1 Schematic flowchart of the soil moisture prediction method provided by an embodiment of the present application;

[0032] Figure 2 Schematic diagram of the Uformer deep learning network structure;

[0033] Figure 3 Schematic diagram of the LeWin Transformer module structure;

[0034] Figure 4 Schematic diagram of the LEFF structure;

[0035] Figure 5 Schematic diagram of the mean square error curve;

[0036] Figure 6 Schematic diagram of the structure of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0038] The purpose of this application is to provide a soil moisture prediction method, device, medium and product, aiming to improve the prediction accuracy of soil moisture.

[0039] To make the above objects, features and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] In an exemplary embodiment, as Figure 1 shown, a soil moisture prediction method is provided, including:

[0041] Step 1: Obtain the target soil moisture time series data; the target soil moisture time series data includes: the true values of the soil moisture at each position in the to-be-predicted area at each historical moment in the historical period.

[0042] Step 2: Input the target soil moisture time series data into the soil moisture prediction model to obtain the predicted values of the soil moisture at each position in the to-be-predicted area at each to-be-predicted moment in the to-be-predicted period.

[0043] Among them, the soil moisture prediction model is obtained by training the Uformer deep learning network using a training set, and the Uformer deep learning network is constructed based on the LeWin Transformer module.

[0044] As an optional implementation manner, in Step 2, the training process of the soil moisture prediction model includes:

[0045] Step 211: Obtain a training set; the training set includes multiple groups of training data, and the training data includes: the true values of the soil moisture at each position in the sample area at each historical moment in two adjacent historical periods.

[0046] Step 212: Construct a Uformer deep learning network based on the LeWin Transformer module.

[0047] As an optional implementation manner, as Figure 2 shown, the Uformer deep learning network includes: an input projection module, an encoder, a bottleneck layer, a decoder, and an output projection module connected in sequence, and the encoder and the decoder are skip-connected.

[0048] As an alternative implementation, the encoder includes a plurality of LeWin Transformer modules and a plurality of downsampling modules, the bottleneck layer includes 1 LeWin Transformer module, the decoder includes a plurality of LeWin Transformer modules and a plurality of upsampling modules, and the number of LeWin Transformer modules and downsampling modules in the encoder is equal to the number of LeWin Transformer modules and upsampling modules in the decoder.

[0049] Specifically, as Figure 3 shown, the LeWin Transformer module includes: 2 layer normalization layers (LN), a windowed self-attention mechanism module (W-MSA), and a locally-enhanced feed-backward (LeFF) network.

[0050] As Figure 4 shown, the LeFF includes: 2 1×1 convolutional (Conv) layers, a flattening layer (Flatten), a 3×3 depth-wise convolutional layer, and a Reshape function.

[0051] Step 213: Use the training set to train the Uformer deep learning network to obtain a soil moisture prediction model.

[0052] As an alternative implementation, step 213 includes:

[0053] Step 2131: Based on the power spectral density and the mean square error, use the true values of the soil moisture at each position in the sample area at the previous historical moment in each group of training data in the training set as the input, and use the true values of the soil moisture at each position in the corresponding sample area at the next historical moment as the output to train the Uformer deep learning network to obtain a soil moisture prediction model.

[0054] Specifically, the training process at any current training iteration in step 2131 specifically includes:

[0055] (1) Input the true values of the soil moisture at each position in the sample area at the previous historical moment in each group of training data in the training set into the Uformer deep learning network at the current training iteration to obtain the predicted values of the soil moisture at each position in the corresponding sample area at the next historical moment at the current training iteration.

[0056] (2) Using the mean square error calculation formula, based on the predicted values of soil moisture at each position in the sample area at the subsequent historical moment of the current training iteration in each group of training data and the corresponding true values of soil moisture at each position in the sample area at the subsequent historical moment, calculate the mean square error at the current training iteration; the mean square error calculation formula is:

[0057]

[0058] where, MSE is the mean square error; N is the total number of groups of training data in the training set; P pred,i is the predicted value of soil moisture at each position in the sample area at the subsequent historical moment of the i-th group of training data; P true,i is the true value of soil moisture at each position in the sample area at the subsequent historical moment of the i-th group of training data.

[0059] (3) Using the KL divergence calculation formula, based on the predicted values of soil moisture at each position in the sample area at the subsequent historical moment of the current training iteration in each group of training data and the corresponding true values of soil moisture at each position in the sample area at the subsequent historical moment, calculate the KL divergence at the current training iteration; the KL divergence calculation formula includes:

[0060]

[0061]

[0062] where, KL is the KL divergence; PSD' pred,i is the normalized power spectral density corresponding to the predicted value of soil moisture at each position in the sample area at the subsequent historical moment of the i-th group of training data; PSD t ' rue,i is the normalized power spectral density corresponding to the true value of soil moisture at each position in the sample area at the subsequent historical moment of the i-th group of training data; PSD pred,i is the original power spectral density corresponding to the predicted value of soil moisture at each position in the sample area at the subsequent historical moment of the i-th group of training data; PSD pred,j is the original power spectral density corresponding to the predicted value of soil moisture at each position in the sample area at the subsequent historical moment of the j-th group of training data; PSD true,i is the original power spectral density corresponding to the true value of soil moisture at each position in the sample area at the subsequent historical moment of the i-th group of training data; PSD true,j is the original power spectral density corresponding to the true value of soil moisture at each position in the sample area at the subsequent historical moment of the j-th group of training data; M pred,i is the amplitude spectrum corresponding to the predicted value of soil moisture at each position in the sample area at the subsequent historical moment of the i-th group of training data; Mtrue,i is the amplitude spectrum corresponding to the true value of the soil humidity at each position in the sample area at the subsequent historical moment in the i-th group of training data; FFT pred,i is the Fourier transform result of the predicted value of the soil humidity at each position in the sample area at the subsequent historical moment in the i-th group of training data; FFT true,i is the Fourier transform result of the true value of the soil humidity at each position in the sample area at the subsequent historical moment in the i-th group of training data; |·| is the absolute value symbol; F(·) is the Fourier transform.

[0063] (4) Using the total loss calculation formula, based on the mean square error and KL divergence at the current training iteration, calculate the total loss at the current training iteration; the total loss calculation formula is:

[0064] LOSS = MSE + KL.

[0065] where LOSS is the total loss.

[0066] (5) Calculate the difference between the total loss at the current training iteration and the total loss at the previous training iteration to obtain the difference at the current training iteration.

[0067] (6) Determine whether the stop condition is satisfied; the stop condition is: the difference at the current training iteration is less than the preset difference or the preset training iteration is reached.

[0068] (7) If so, determine the Uformer deep learning network at the current training iteration as the soil humidity prediction model.

[0069] (8) If not, adjust the parameters of the Uformer deep learning network at the current training iteration, and return "input the true value of the soil humidity at each position in the sample area at the previous historical moment in each group of training data in the training set into the Uformer deep learning network at the current training iteration to obtain the predicted value of the soil humidity at each position in the sample area at the subsequent historical moment corresponding to the current training iteration", until the stop condition is satisfied to obtain the soil humidity prediction model.

[0070] As an optional implementation manner, after step 213, it further includes:

[0071] Use the mean square error to evaluate the performance of the soil humidity prediction model.

[0072] As an optional implementation manner, step 2 includes:

[0073] Step 221: Input the target soil humidity time series data into the input projection module in the soil humidity prediction model for convolution operation to obtain the initial features of the area to be predicted.

[0074] Step 222: Input the initial features of the area to be predicted into the encoder in the soil moisture prediction model for multiple windowed self-attention operations, local enhancement, and downsampling processing to obtain the encoded features of the area to be predicted.

[0075] Specifically, at each stage of the encoder, the initial features pass through a set of LeWin Transformer modules and, in conjunction with the downsampling layer, gradually reduce the spatial resolution and increase the number of channels to form multi-scale moisture features, i.e., encoded features. The LeWin Transformer module uses W-MSA to capture long-range dependencies and reduces the computational complexity through non-overlapping windows, making the processing of large-scale soil moisture data more efficient.

[0076] Specifically, the formula for the windowed self-attention operation and local enhancement of the input features by the g-th LeWin Transformer module (abbreviated as the LeWin module) in the encoder is:

[0077] X' g = WMSA(LN(X g-1 )) + X g-1 .

[0078] X g = LEFF(LN(X' g )) + X' g .

[0079] Among them, X' g is the intermediate feature of the g-th LeWin Transformer module; WMSA(·) is the windowed self-attention operation of W-MSA; LN(·) is the layer normalization operation of the layer normalization layer; X g-1 is the feature output by the (g - 1)-th LeWin Transformer module; X g is the feature output by the g-th LeWin Transformer module; LEFF(·) is for local enhancement by LeFF.

[0080] Step 223: Input the encoded features of the area to be predicted into the bottleneck layer in the soil moisture prediction model for windowed self-attention operation and local enhancement processing to obtain the features after bottleneck layer processing of the area to be predicted.

[0081] Specifically, the bottleneck layer is located between the encoder and the decoder and is responsible for compressing the features and global semantic aggregation to extract longer-range moisture change information. Through the bottleneck layer, the model can learn a wider moisture distribution and dynamics, and thus provide richer feature inputs for the decoder.

[0082] Step 224: Input the features processed by the bottleneck layer of the area to be predicted into the decoder in the soil moisture prediction model for multiple windowed self-attention operations, local enhancement, and upsampling processing to obtain the decoded features of the area to be predicted.

[0083] Specifically, the upsampling layer gradually restores the spatial resolution of the feature map through transposed convolution operations (inverse convolution), and performs feature fusion with the corresponding layers of the encoder through skip connections at each stage to ensure the consistency of local details and global semantics.

[0084] Step 225: Input the decoded features of the area to be predicted into the output projection module in the soil moisture prediction model for convolution operations to obtain the predicted values of the soil moisture at each position in the area to be predicted at each prediction moment during the period to be predicted.

[0085] Furthermore, to verify the method of this application, the mean squared errors (MSE) of soil moisture prediction at three different depths of 0 - 10 cm, 10 cm - 40 cm, and 40 cm - 100 cm were also evaluated, and the evaluation period was from the 1st day to the 7th day in the future. Table 1 details the mean squared errors of soil moisture prediction at different depths. The corresponding mean squared error curve of Table 1 is as Figure 5 shown.

[0086] Table 1 Mean Squared Error Table

[0087]

[0088]

[0089] Based on Table 1 and Figure 5 the following conclusions can be obtained:

[0090] 1) Prediction of surface soil moisture at 0 - 10 cm.

[0091] In the surface soil, the MSE increases significantly day by day, from 0.0259 on the 1st day to 0.1519 on the 7th day. This indicates that as the prediction time increases, the error accumulates continuously and the prediction accuracy gradually decreases. The surface soil moisture is more easily affected by external factors such as precipitation, temperature, and wind speed, and the prediction difficulty is relatively high.

[0092] This increasing trend of error shows that the soil moisture prediction model is more interfered in the prediction of surface soil moisture, especially in the prediction of longer time scales, and it may be difficult to fully capture the short-term drastic changes of the surface soil.

[0093] 2) Prediction of middle soil moisture at 10 cm - 40 cm.

[0094] The prediction error of the middle-layer soil increases from 0.0087 on the first day to 0.0648 on the seventh day. Although the growth rate is obvious, the overall error is relatively smaller than that of the surface layer. As the number of prediction days increases, the MSE value shows a stable growth trend, but there is no drastic change as in the case of the surface layer.

[0095] This indicates that the humidity of the middle-layer soil is relatively less affected by the surface layer and has a more stable humidity change pattern. The soil humidity prediction model is relatively more accurate in capturing the humidity changes in this layer.

[0096] 3) Prediction of soil humidity in the deep layer of 40 cm - 100 cm.

[0097] The prediction error of the deep-layer soil is the smallest, increasing from 0.0034 on the first day to 0.0503 on the seventh day. This trend shows that the humidity change of the deep-layer soil is the most stable. The error of the soil humidity prediction model increases slowly when predicting the humidity of the deep-layer soil, and the prediction performance is relatively the most stable.

[0098] Since the humidity of the deep-layer soil is relatively less affected by external environmental changes, the soil humidity prediction model is easier to learn and fit the humidity changes in the deep layer, which makes the prediction results of the deep-layer soil have higher reliability and stability.

[0099] Overall trend of error: Generally, the prediction error increases over time, which is a common phenomenon in time series prediction models because long-term prediction will lead to the accumulation of errors.

[0100] 4) Differences between the surface layer and the deep layer.

[0101] The error growth of the surface-layer soil humidity is the most significant, especially during the period from the fifth day to the seventh day, with a large increase in error. The humidity of the surface-layer soil is affected by multiple external factors such as precipitation and evaporation, and has strong volatility, which increases the difficulty of prediction.

[0102] The prediction performance of the deep-layer soil humidity is the most stable, with a slow growth of error, which indicates that the humidity change of the deep-layer soil is relatively small, and the soil humidity prediction model can better grasp the change characteristics of the deep-layer humidity.

[0103] In an exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the soil humidity prediction method.

[0104] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the soil humidity prediction method is implemented.

[0105] In an exemplary embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements a soil moisture prediction method.

[0106] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structural diagram can be as Figure 6 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements a soil moisture prediction method.

[0107] Those skilled in the art can understand that Figure 6 the structure shown in

[0108] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memories can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0109] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0110] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0111] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the methods and core ideas of the present application; at the same time, for those of ordinary skill in the art, according to the ideas of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A soil moisture prediction method, characterized in that: The soil moisture prediction method comprises: Acquire target soil moisture time series data; the target soil moisture time series data includes: the true value of soil moisture at each location of the area to be predicted at each historical moment in the historical period; The target soil moisture time series data is input into the soil moisture prediction model to obtain the predicted value of soil moisture at each location in the predicted area at each predicted time in the predicted period; the soil moisture prediction model is obtained by using the training set and training the Uformer deep learning network, and the Uformer deep learning network is built based on the LeWinTransformer module.

2. The soil moisture prediction method according to claim 1, characterized in that: The training process of the soil moisture prediction model includes: Acquire the training set; the training set includes multiple groups of training data, and the training data includes: the true value of soil moisture at each location of the sample area at each historical moment in two adjacent historical periods; Construct the Uformer deep learning network based on the LeWin Transformer module; The Uformer deep learning network is trained using the training set to obtain the soil moisture prediction model.

3. The soil moisture prediction method according to claim 2, characterized in that: The Uformer deep learning network includes: an input projection module, an encoder, a bottleneck layer, a decoder and an output projection module connected in sequence, and the encoder and the decoder are jump-connected.

4. The soil moisture prediction method according to claim 3, characterized in that: The encoder includes multiple LeWinTransformer modules and multiple downsampling modules, the bottleneck layer includes 1 LeWin Transformer module, the decoder includes multiple LeWin Transformer modules and multiple upsampling modules, and the number of LeWin Transformer modules and downsampling modules in the encoder and the number of LeWin Transformer modules and upsampling modules in the decoder are equal.

5. The soil moisture prediction method according to claim 4, characterized in that: The target soil moisture time series data is input into the soil moisture prediction model to obtain the predicted value of soil moisture at each location in the to-be-predicted area at each to-be-predicted time in the to-be-predicted period, including: Inputting the target soil moisture time series data into the input projection module in the soil moisture prediction model for convolution operation to obtain the initial features of the area to be predicted; The initial features of the area to be predicted are input into the encoder of the soil moisture prediction model for multiple windowed self-attention operations, local enhancement and downsampling processing to obtain the encoding features of the area to be predicted; The encoded features of the area to be predicted are input into the bottleneck layer of the soil moisture prediction model for windowed self-attention operation and local enhancement processing to obtain the features of the area to be predicted after bottleneck layer processing; The features processed by the bottleneck layer of the area to be predicted are input into the decoder in the soil moisture prediction model for multiple windowed self-attention operations, local enhancement and upsampling processing to obtain the decoded features of the area to be predicted; The decoded features of the area to be predicted are input into the output projection module in the soil moisture prediction model for convolution operation to obtain the predicted value of soil moisture at each position of the area to be predicted at each predicted time in the predicted period.

6. The soil moisture prediction method according to claim 3, characterized in that: The Uformer deep learning network is trained using the training set to obtain the soil moisture prediction model, including: Based on power spectral density and mean square error, the true value of soil moisture at each location in the sample area at the previous historical moment in each group of training data in the training set is used as input, and the true value of soil moisture at each location in the sample area at the corresponding next historical moment is used as output, the Uformer deep learning network is trained to obtain the soil moisture prediction model.

7. The soil moisture prediction method according to claim 2, characterized in that: After the Uformer deep learning network is trained using the training set to obtain the soil moisture prediction model, the method further includes: The mean square error was used to evaluate the performance of the soil moisture prediction model.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the soil moisture prediction method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the soil moisture prediction method according to any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the soil moisture prediction method according to any one of claims 1 to 7 is implemented.

Images