Method for realizing soil humidity prediction based on LSTM model of meta learning

Through the LSTM model based on meta-learning, the problem of spatiotemporal heterogeneity in the global soil moisture prediction of traditional deep learning methods is solved, the adaptability to different climate regions is improved, and the prediction accuracy is significantly improved, especially in cold zones and winter arid areas. .

CN119066952AInactive Publication Date: 2024-12-03CHANGCHUN NORMAL UNIV

Patent Information

Application Number
CN202411044999.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-12-03
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional deep learning methods perform inconsistently when dealing with spatiotemporal heterogeneity of global soil moisture, making it difficult to accurately predict soil moisture changes in some areas, increasing the difficulty of global soil moisture prediction.

Method used

The LSTM model based on meta-learning is adopted to divide the meta-learning areas through Coben climate classification, and train with LandBench data and DEM data to realize meta-training and meta-updation, enhancing the model's adaptability to different climate characteristics.

Benefits of technology

The model's learning ability to hard-to-predict areas has been improved, the ability to generalize soil moisture changes worldwide has been enhanced, and the accuracy of global soil moisture prediction has been significantly improved, especially in cold zones and winter arid areas.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066952A_ABST
    Figure CN119066952A_ABST
Patent Text Reader

Abstract

The invention provides a method for realizing soil humidity prediction based on an LSTM model of meta learning, and the method comprises the following steps: the Kohen climate classification in input data is used for dividing meta training and meta updating regions in the meta learning, LandBench data and DEM are the input training data of the model, the meta learning thought is respectively used in LSTM, EDLSTM and AttLSTM models, and the soil humidity of the next day is predicted. And carrying out comprehensive evaluation on all models to complete prediction. According to the method, the learning ability of the model for an area difficult to predict is improved, so that the generalization ability of the model for adapting to soil humidity changes in the global range is enhanced, training is performed in a LandBench data set, the result shows that compared with different LSTM-based models, the global soil humidity performance is obviously improved, and the global soil humidity performance is improved in a frigid zone and an arid region in winter. A model which is not integrated with a meta-learning structure is difficult to predict. And by integrating the meta-learning architecture, the global prediction precision is improved, and the prediction performance of the regions is also remarkably enhanced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of soil moisture prediction, and in particular relates to a soil moisture prediction method based on an LSTM model based on meta-learning. Background Art

[0002] As the foundation of terrestrial ecosystems, soil moisture conditions are crucial to maintaining ecological balance and the normal operation of the climate system. Soil moisture not only directly affects the water and heat flux exchange between the ground and the atmosphere, but its changes also play an important role in weather patterns and long-term climate change. In addition, the dynamic changes in soil moisture also have memory, which can record and feedback abnormal signals in the atmosphere, further affecting the spatiotemporal evolution of climate and ecosystems. Therefore, accurate prediction of soil moisture is extremely important for a deep understanding of climate change and ecosystem dynamics. In recent years, with the development of deep learning technology, long short-term memory networks (LSTM) and their variant models have shown great potential in improving the accuracy of soil moisture prediction. Some researchers have improved the prediction of regional soil moisture by integrating feature attention mechanisms, combining generative adversarial networks, and adopting cross-source transfer learning.

[0003] However, traditional deep learning methods use a homogeneous learning model, that is, gradient descent based on the average loss of input data, which leads to inconsistent performance of the model when dealing with the spatiotemporal heterogeneity of global soil moisture because the relationship between different regions is complex and changeable. This phenomenon makes it easy for deep learning models to simulate soil moisture changes in some regions, but in other regions, the prediction accuracy is low, increasing the difficulty of global soil moisture prediction.

[0004] Based on this, a meta-learning-based LSTM model is proposed to realize soil moisture prediction method. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a soil moisture prediction method based on an LSTM model based on meta-learning in order to solve the problems raised in the above-mentioned background technology.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: a soil moisture prediction method is implemented based on a meta-learning LSTM model, including the following steps: the Köppen climate classification in the input data is used to divide the meta-training and meta-update areas in the meta-learning, LandBench data and DEM are the input training data of the model, and the meta-learning idea is used in the LSTM, EDLSTM, and AttLSTM models respectively to predict the soil moisture of the next day. All models are comprehensively evaluated to complete the prediction.

[0007] As a further illustration of the present invention, the meta-learning model fθ It includes two main stages: meta-training and meta-updating. In the meta-training stage, the meta-learning model is copied, the copied model training data is used, the parameters are updated, and finally the intermediate parameters are obtained. renew The formula is as follows;

[0008]

[0009] Among them, α is the learning rate,

[0010] T i is the i-th task randomly selected from the task distribution p(T),

[0011] is the gradient on the i-th task;

[0012] Then use the model trained in the meta-training phase, that is, use the parameters Perform training in the meta-update phase, accumulate the loss obtained in the meta-update phase to update the meta-learning model f θ The parameter θ, the formula for updating θ is as follows;

[0013]

[0014] in, is the model trained in the meta-training phase,

[0015] is the loss of the cumulative meta-update phase, which eventually updates the parameters of the meta-learning model;

[0016] In the next cycle, the meta-training model copies the updated meta-learning model again. The meta-update phase uses the parameters obtained in the meta-training phase. The meta-training phase does not update the meta-learning model but uses the meta-learning model updated in the meta-update phase. This cycle is used to train the model to adapt to different climate characteristics.

[0017] As a further illustration of the present invention, the evaluation criteria selected R 2 , KGE and RMSE, R 2 It is an indicator to measure the model's ability to explain data variability; KGE is an indicator that comprehensively evaluates the similarity, deviation and time series characteristics between predictions and observations; RMSE is a model performance evaluation indicator widely used in meteorological, hydrological and climate research.

[0018] As a further illustration of the present invention,

[0019]

[0020]

[0021] y i and x i Represent the actual value and predicted value of the i-th time step respectively;

[0022] and are the averages of the actual and predicted values, respectively;

[0023] r is the linear correlation coefficient between the actual value and the predicted value;

[0024] β is the proportionality coefficient between the actual value and the predicted value;

[0025] γ is the time series correlation coefficient between the actual value and the predicted value;

[0026] The minimum-maximum normalization method is used to scale the data to the range of [0,1], which not only retains the relative relationship between the data but also makes it easy to understand. The formula is as follows;

[0027]

[0028] Among them, x, x max , x min , x norm are the original value, maximum value, minimum value and normalized value of the training data on the grid respectively.

[0029] Compared with the prior art, the present invention has the following advantages:

[0030] The soil moisture prediction method based on meta-learning LSTM model in the present invention improves the model's learning ability for difficult-to-predict areas, thereby enhancing its generalization ability to adapt to soil moisture changes on a global scale. The training results on the LandBench dataset show that compared with different LSTM-based models, the global soil moisture performance is significantly improved. Specifically, the median R of the predicted 1-day 2 The prediction accuracy of the model without the meta-learning structure was improved by 2.05%-7.54%, the KGE was improved by 1.84%-5.50%, and the RMSE was improved by 4.44%-12.50%. In the cold zone and winter drought areas, the prediction of the model without the meta-learning structure was difficult. By incorporating the meta-learning architecture, the global prediction accuracy was improved, and the prediction performance in these areas was significantly enhanced, providing a new paradigm for global soil moisture simulation based on deep learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a framework diagram of the present invention using a meta-learning model to predict soil moisture;

[0032] Figure 2 In the embodiment of the present invention, R 2The indicator evaluates the prediction performance of various deep learning models for soil moisture at 1 degree spatial resolution and 1 day lead time.

[0033] Figure 3 is a framework diagram of soil moisture prediction in which the evaluation index is RMSE in an embodiment of the present invention;

[0034] Figure 4 It is an analysis diagram of the CDF graphs of all models in the embodiment of the present invention;

[0035] Figure 5 is a box plot of the prediction performance of EDLSTM, EDLSTM (Meta-R12), AttLSTM and AttLSTM (Meta-R12) models in soil moisture prediction for each climate zone in an embodiment of the present invention;

[0036] Figure 6 It is a scatter plot of soil moisture prediction and observation by AttLSTM (a, c, e, g, i) and AttLSTM (b, d, f, h, j) with meta-learning model in the embodiment of the present invention;

[0037] Figure 7 In the embodiment of the present invention, Figure 5 The same models are LSTM, LSTM(Meta-R12) and LSTM(Transfer-R12), and the evaluation metric is KGE.

[0038] Figure 8 A one-year time series of soil moisture at a randomly selected grid on the global map and (ae) at different grids.

[0039] Fig. 9 1 is a diagram of an experimental setup of an LSTM model in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] like Figure 1As shown, the present invention provides a technical solution: a soil moisture prediction method based on a meta-learning LSTM model, comprising the following steps: the Köppen climate classification in the input data is used to divide the meta-training and meta-update areas in the meta-learning, the LandBench data and the DEM are the input training data of the model, and the meta-learning idea is used in the LSTM, EDLSTM, and AttLSTM models respectively to predict the soil moisture for the next day. All models are comprehensively evaluated to complete the prediction.

[0042] Meta-learning model f θ It includes two main stages: meta-training and meta-updating. In the meta-training stage, the meta-learning model is copied, the copied model training data is used, the parameters are updated, and finally the intermediate parameters are obtained. renew The formula is as follows;

[0043]

[0044] Among them, α is the learning rate,

[0045] T i is the i-th task randomly selected from the task distribution p(T),

[0046] is the gradient on the i-th task;

[0047] Then use the model trained in the meta-training phase, that is, use the parameters Perform training in the meta-update phase, accumulate the loss obtained in the meta-update phase to update the meta-learning model f θ The parameter θ, the formula for updating θ is as follows;

[0048]

[0049] in, is the model trained in the meta-training phase,

[0050] is the loss of the cumulative meta-update phase, which eventually updates the parameters of the meta-learning model;

[0051] In the next cycle, the meta-training model copies the updated meta-learning model again. The meta-update phase uses the parameters obtained in the meta-training phase. The meta-training phase does not update the meta-learning model but uses the meta-learning model updated in the meta-update phase. This cycle is used to train the model to adapt to different climate characteristics.

[0052] The evaluation criteria selected R 2 , KGE and RMSE, R 2It is an indicator to measure the model's ability to explain data variability; KGE is an indicator that comprehensively evaluates the similarity, deviation and time series characteristics between predictions and observations; RMSE is a model performance evaluation indicator widely used in meteorological, hydrological and climate research.

[0053]

[0054] y i and x i Represent the actual value and predicted value of the i-th time step respectively;

[0055] and are the averages of the actual and predicted values, respectively;

[0056] r is the linear correlation coefficient between the actual value and the predicted value;

[0057] β is the proportionality coefficient between the actual value and the predicted value;

[0058] γ is the time series correlation coefficient between the actual value and the predicted value;

[0059] The minimum-maximum normalization method is used to scale the data to the range of [0,1], which not only retains the relative relationship between the data but also makes it easy to understand. The formula is as follows;

[0060]

[0061] Among them, x, x max , x min , x norm are the original value, maximum value, minimum value and normalized value of the training data on the grid respectively.

[0062] Through a large number of experimental examples, this embodiment determines a set of optimized long short-term memory (LSTM) model parameters.

[0063] Specifically, the learning rate is set to 0.001 to promote stable convergence of the model during training; 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 dropout rate is set to 0.15 to prevent the model from being overly sensitive to any single feature or pattern in the training data; finally, the sequence length is 365 days, which enables the model to recognize seasonal changes and long-term dependencies throughout the year.

[0064] In the experimental example, the LSTM model is used to predict the global soil moisture for the next day to verify the reliability of the deep learning model in hydrological prediction. Subsequently, the LSTM (Meta-R9) model, LSTM (Meta-R12) model and LSTM (Meta-R9+12) model are used to make the same prediction. It can be seen that the Meta-R12 model has the best effect. In order to test whether the meta-learning idea can be integrated into other models, two more sets of experimental examples are made to ensure that the experimental example parameters are the same. The same seed is taken to use the EDLSTM model and EDLSTM (Meta R12) model and the AttLSTM model and AttLSTM (Meta-R12) model to make the same prediction.

[0065] Figure 2 is to use R 2 The indicators evaluate the prediction performance of various deep learning models for soil moisture at 1 degree spatial resolution and 1 day lead time. The results are shown in the first row for the LSTM model, the second row for the EDLSTM model, and the third row for the AttLSTM model. The first column shows the prediction model without meta-learning, the second column shows the prediction model with meta-learning applied, and the third column shows the improvement after meta-learning integration. The yellow and red shades in the third column indicate the performance improvement after meta-learning, while blue indicates a performance decrease. The color intensity reflects the degree of improvement or decrease. Meta-R12 represents the meta-update stage of the meta-learning model trained on the 12th region data described in Table 1.

[0066] Table 1 Climate zone codes and descriptions, as well as data quantity and prediction evaluation indicators

[0067]

[0068] The bold parts represent the regions considered in the meta-update phase before identifying the best performing meta-learning model for global soil moisture prediction.

[0069] Figure 3 is with Figure 1 Same, the evaluation index is RMSE;

[0070] Figure 4 The CDF graphs of all models are analyzed for R2(a, b), KGE(c, d) and RMSE(e, f) indicators respectively. The dark blue dashed line indicates the focus on local magnification. Meta-R12 indicates the meta-update stage of the meta-learning model in the 12th region training data described in Table 1.

[0071] Figure 5Figure 1 is a box plot of the prediction performance of the EDLSTM, EDLSTM(Meta-R12), AttLSTM, and AttLSTM(Meta-R12) models in soil moisture prediction for each climate zone, measured using R2. In addition, a gray line graph shows the number of data at 1° resolution for all climate zones. Meta-R12 indicates that the meta-update stage of the meta-learning model trains data in the 12th zone. The description of R1-R12 is shown in Table 1.

[0072] Figure 6 Scatter plots of predicted and observed soil moisture by AttLSTM (a, c, e, g, i) and AttLSTM with meta-learning model (b, d, f, h, j).

[0073] The first row shows the situation globally, while the subsequent rows describe the situation in regions R3, R4, R8, and R10 in detail. Meta-R12 means that the meta-update stage of the meta-learning model trains data in region 12, as described in Table 1.

[0074] In terms of spatial patterns, the three indicators of R2, KGE and NSE in the LSTM model prediction results show similar states, and the prediction results in northern North America, northern Africa and central Asia are poor. The purple circle is the result of the LSTM model prediction, and the blue circle is the result of the Meta-R12 prediction. Generally speaking, regions with higher correlation coefficients have lower root mean square errors, and vice versa. Therefore, the spatial distribution map of RMSE is opposite to that of the other three indicators. It can be seen from the green circle that the Meta-R12 model has made great improvements in these places, especially for the indicator KGE. The climate types in northern North America, northern Africa and central Asia are different, but the meta-learning model can improve the prediction effect, proving that the meta-learning model has good generalization performance compared with the traditional LSTM model. However, the prediction performance of some areas in southwestern China, especially in Yunnan, China, and central Australia (in the yellow circle) has declined. The climate in Yunnan is diverse, mainly dominated by subtropical monsoon climate and tropical monsoon climate, and the high precipitation in summer may be the reason for affecting soil moisture prediction. Australia has mostly semi-arid and desert climates, and the environment is complex and changeable, which may lead to reduced prediction performance.

[0075] Analysis of regional characteristics of different models

[0076] The Köppen climate classification system divides regions into 13 regions according to the secondary divisions. The 13th region covers most of the Antarctic Circle and Greenland. These areas are covered by ice and snow all year round, and it is common for the soil to be exposed. They are often affected by extreme low temperatures and frozen ground. Due to the lack of soil moisture data, these areas were not included in the analysis.

[0077] Table 2 Prediction results of all models.

[0078]

[0079] The bold parts indicate the better performance between the model with meta-learning and the model without meta-learning.

[0080] This experiment uses 1° resolution data and the number of grid points in each partition as the second coordinate. Based on the model based on the long short-term memory network (LSTM), area 9 is used as the meta-update area, area 12 is used as the meta-update area, and area 9 and area 12 are used as the meta-update area. The evaluation index is R2. The results are shown in Figure 7 As shown in the box plot results, except for area 1 and area 12, the prediction results of other areas have improved, indicating that incorporating meta-learning ideas into deep learning models is beneficial for soil moisture prediction.

[0081] Randomly selected grids on the global map and (ae) soil moisture time series for one year in different grids. These time series are generated using the AttLSTM model and the AttLSTM with meta-learning model, which performs best among all models, with a lead time of 1 day. Meta-R12 means that the meta-update stage of the meta-learning model is trained on the 12th region data, and the specific description is shown in Table 1.

[0082] Region 1 has a tropical rainforest climate, while region 12 has a polar climate. Since the generalized features learned by meta-learning on global regions may not be applicable to extreme climates, they perform poorly in these climate conditions.

[0083] Region 6 is temperate and dry in summer. There are few experimental grid points, and the Meta-R12 model works well, indicating that meta-learning can improve data sparsity (Luo et al., 2022). Region 9 is in the frigid zone and dry in summer. It is mainly distributed in northern Russia and northern Canada. The terrain is high, the temporal and spatial heterogeneity is strong, and data collection is difficult. The prediction effect of traditional LSTM for region 9 is not ideal, and the Meta-R12 model has a significant improvement in predicting region 9. Region 12 belongs to the polar region and tundra, region 6 belongs to the temperate zone with dry summer climate, and region 9 belongs to the cold zone with dry summer climate. Summer drought may play a protective role for permafrost because dry soil has stronger thermal insulation properties, and due to reduced rainwater infiltration, heat input is also reduced accordingly (Magnússon et al., 2022). It may be because these regions may experience obvious seasonal changes and strong temporal and spatial heterogeneity. Using region 12 as the meta-update stage will greatly improve the prediction performance of regions 6 and 9.

[0084] In addition, in order to compare the learning effects of meta-learning and transfer learning, the data of the meta-training phase of the Meta-R12 model is used as the data of the pre-training phase of the Transfer-LSTM model, and the data of the meta-update phase of the Meta-R12 model is used as the data of the fine-tuning phase of the Transfer-LSTM model. Keeping other hyperparameters unchanged, the soil moisture value of the next day is also predicted, and the final model KGE value is obtained.

[0085] Figure 8 is a one-year time series of soil moisture in a randomly selected grid on the global map and (ae) in different grids. The time series are generated using the AttLSTM model and the AttLSTM with meta-learning model (the best performance among all models) with a lead time of 1 day. Meta-R12 represents the data in the training region 12 in the meta-update phase of the meta-learning model, as shown in Table 1

[0086] Fig. 9 It is an experimental device for the LSTM model. Lon, Lat, F, and T represent the data size of four dimensions: longitude, latitude, feature, and time.

[0087] Transfer learning has a slight improvement in area 12, but the model performance in other areas and even in the overall area has dropped significantly. Meta-learning has a significant improvement in most areas and the overall area.

[0088] For the time characteristics analysis, several coordinate points in different regions of the world were selected for time series analysis. These points are located in Germany (a), Argentina (b), the United States (c), the Democratic Republic of the Congo (d) and Russia (e). It can be clearly observed that the prediction curve of the Meta-R12 model (marked in the red circle) is more accurate than the prediction curve of the LSTM model alone, and can even be closely aligned with the observed values. From these charts, it can be seen that the Meta-R12 model can effectively capture the changes in soil moisture regardless of whether the soil moisture value increases or decreases. Even in locations in different countries with large differences in geomorphological features, the meta-learning model can more accurately predict the value of soil moisture than using only the LSTM model, which also proves the rationality of incorporating the idea of ​​meta-learning into the LSTM model.

[0089] The blue circle is located in Schleswig-Holstein, Germany. This region is located in northern Germany. The landforms are usually relatively flat and may include farmland, forests, and some small hills. The cultivated area is large, and the soil moisture value may be affected by irrigation, so the prediction performance is poor (Fang & Shen, 2020). Due to the complexity of the landform characteristics and environmental background, the meta-learning model may not be able to effectively learn useful generalized features, so it performs poorly in soil moisture prediction.

[0090] During periods of significant soil moisture fluctuations, precipitation is one of the main factors causing soil moisture fluctuations (Yang et al., 2018). In particular, the Meta-R12 model showed excellent prediction capabilities when soil moisture changed significantly or remained in a dynamic equilibrium state. These results highlight the powerful performance of the meta-learning-based LSTM model in simulating the relationship between soil moisture and surface water, providing meaningful value for the prediction of soil moisture.

[0091] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0092] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A soil moisture prediction method based on a meta-learning LSTM model is implemented, which is characterized by: The following steps are involved: The Köppen climate classification in the input data is used to divide the meta-training and meta-update areas in meta-learning. LandBench data and DEM are the input training data of the model. The meta-learning idea is used in the LSTM, EDLSTM, and AttLSTM models to predict soil moisture for the next day, and all models are comprehensively evaluated to complete the prediction.

2. The method for predicting soil moisture using the LSTM model based on meta-learning according to claim 1, characterized in that: Meta-learning model f θ It includes two main stages: meta-training and meta-updating. In the meta-training stage, the meta-learning model is copied, the copied model training data is used, the parameters are updated, and finally the intermediate parameters are obtained. renew The formula is as follows; Among them, α is the learning rate, T i is the i-th task randomly selected from the task distribution p(T), is the gradient on the i-th task; Then use the model trained in the meta-training phase, that is, use the parameters Perform training in the meta-update phase, accumulate the loss obtained in the meta-update phase to update the meta-learning model f θ The parameter θ, the formula for updating θ is as follows; in, is the model trained in the meta-training phase, is the loss of the cumulative meta-update phase, which eventually updates the parameters of the meta-learning model; In the next cycle, the meta-training model copies the updated meta-learning model again. The meta-update phase uses the parameters obtained in the meta-training phase. The meta-training phase does not update the meta-learning model but uses the meta-learning model updated in the meta-update phase. This cycle is used to train the model to adapt to different climate characteristics.

3. The method for predicting soil moisture using the LSTM model based on meta-learning according to claim 2, characterized in that: The evaluation criteria selected R 2 , KGE and RMSE, R 2 It is an indicator to measure the model's ability to explain data variability; KGE is an indicator that comprehensively evaluates the similarity, deviation and time series characteristics between predictions and observations; RMSE is a model performance evaluation indicator widely used in meteorological, hydrological and climate research.

4. The method for predicting soil moisture using a LSTM model based on meta-learning according to claim 3, characterized in that: y i and x i Represent the actual value and predicted value of the i-th time step respectively; and are the averages of the actual and predicted values, respectively; r is the linear correlation coefficient between the actual value and the predicted value; β is the proportionality coefficient between the actual value and the predicted value; γ is the time series correlation coefficient between the actual value and the predicted value; The minimum-maximum normalization method is used to scale the data to the range of [0,1], which not only retains the relative relationship between the data but also makes it easy to understand. The formula is as follows; Among them, x, x max , x min , x norm are the original value, maximum value, minimum value and normalized value of the training data on the grid respectively.

Citation Information

Patent Citations

  • Mountainous area soil thickness prediction method based on clustering sampling

    CN111275072A

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

    CN114386332A

  • Small sample time sequence prediction method based on meta-learning framework

    CN117272040A

  • Soil humidity prediction system based on improved particle swarm optimization and LSTM

    CN118410828A

Cited By

  • Room temperature light supplement and temperature control device for seedling cultivation based on controllable light intensity and light period

    CN120436006A

  • Material dryness and humidity prediction method and device based on discrete element method and data driving

    CN120510978A