Wheat soil moisture prediction method based on mixed PPLSTM model

Abstract

The invention discloses a wheat soil moisture prediction method based on a mixed PPLSTM model, and the method comprises the following steps: S1, obtaining soil moisture data in a wheat growth period, and carrying out the preliminary screening of data features; s2, performing data preprocessing on the acquired soil moisture data; s3, performing dimension reduction processing on the high-dimensional features by using principal component analysis (PCA); s4, optimizing hyper-parameters of the long short-term memory (LSTM) network by using particle swarm optimization (PSO); s5, inputting the optimized LSTM network into training data for training to obtain a prediction model; s6, testing the trained model by using the verification set, and outputting a prediction result; and S7, optimizing parameters of the LSTM model according to a test result, and performing final training to obtain a model with optimal prediction precision. According to the method, the PSO optimization algorithm, the PCA dimension reduction algorithm and the LSTM network are combined, so that the problems that in a traditional method, high-dimensional input features are redundant, the time dependence relation capturing capacity is insufficient, and hyper-parameter selection depends on experience are solved. Compared with a traditional LSTM model, the method has the advantages that on the basis of keeping the sequence modeling capability, the prediction precision and stability are remarkably improved, and a more reliable decision basis can be provided for intelligent irrigation scheduling, so that the water-saving irrigation practice is effectively promoted, and the sustainable development of precision agriculture is promoted.

Description

Technical Field

[0001] This invention relates to the field of wheat soil moisture prediction technology, and in particular to a method for predicting wheat soil moisture based on a hybrid PPLSTM model. Background Technology

[0002] In wheat production management, precise regulation of soil moisture has a crucial impact on crop growth and safety. It not only directly affects wheat yield but also influences water use efficiency, thereby affecting the stability of grain production and the sustainable use of agricultural resources. The development of deep learning technology has brought new opportunities for soil moisture prediction. Among them, prediction models based on LSTM time-series data, with their high-efficiency processing capabilities, have become an important technical support for realizing intelligent irrigation decision-making.

[0003] However, building efficient soil moisture prediction models in actual farmland environments still faces a series of severe challenges. Traditional prediction methods, when dealing with multi-source, high-dimensional soil and environmental characteristics, often suffer from low model training efficiency and overfitting problems due to redundant and complex correlations of input variables, making it difficult to extract the key factors that truly influence moisture changes. Meanwhile, while prediction models such as LSTM can effectively capture temporal dependencies, their hyperparameter settings significantly impact prediction performance. Traditional manual hyperparameter tuning methods are not only time-consuming and laborious but also struggle to achieve an optimal balance between convergence speed and prediction accuracy, limiting the reliability and practicality of these models in real-time irrigation decision-making.

[0004] Patent document CN119357606A discloses a soil moisture prediction method based on external climate environment. This method uses a grey model to process historical data and generate predicted values ​​for future time points. Then, it uses an LSTM model to correct the predicted values ​​and output the final predicted value. However, this method directly inputs raw data with high feature dimensionality and inherent correlation into the LSTM model, which not only increases the computational complexity of the model but also makes it easier to introduce noise, making it difficult for the model to focus on the most critical climate and soil moisture change patterns, thus affecting the robustness and efficiency of the prediction. Patent document CN118940160A discloses a soil moisture prediction method based on a physical process-integrated attention-based encoding-decoding LSTM model. This method uses the HBV hydrophysical model to generate physically meaningful intermediate variables from raw meteorological and land surface data, and selects key features through correlation analysis. Then, these physical features and raw data are input into an encoding-decoding LSTM model with an integrated attention mechanism for training, and finally, the model is used to predict future surface soil moisture. However, this method uses an attention-based encoder-decoder LSTM model with a huge parameter space. When trained with fixed learning rates and initial weights, it is prone to getting stuck in local optima, resulting in slow convergence and unstable training. This increases the overall computational cost of the system and reduces prediction accuracy.

[0005] Therefore, the present invention provides a new solution to this problem. Summary of the Invention

[0006] In view of the above situation and to overcome the shortcomings of the existing technology, the purpose of this invention is to provide a wheat soil moisture prediction method based on a combined PPLSTM model, so as to solve the above problems existing in the prior art. The specific solution is as follows:

[0007] A method for predicting wheat soil moisture based on a combined PPLSTM model includes the following steps:

[0008] S1. Obtain soil moisture data during the wheat growth cycle and perform preliminary screening of data characteristics;

[0009] S2. Perform data preprocessing on the acquired soil moisture data;

[0010] S3. Use Principal Component Analysis (PCA) to reduce the dimensionality of high-dimensional features;

[0011] S4. Apply particle swarm optimization (PSO) to optimize the hyperparameters of the Long Short-Term Memory (LSTM) network;

[0012] S5. Input the optimized LSTM network into the training data to train it and obtain the prediction model;

[0013] S6. Test the trained model using the validation set and output the prediction results;

[0014] S7. Optimize the parameters of the LSTM model based on the test results, and perform final training to obtain the model with the best prediction accuracy.

[0015] Preferably, the data preprocessing in step S2 includes standardization processing, the specific steps of which are as follows:

[0016] S21. Standardize each feature of the input data so that its mean is 0 and its standard deviation is 1.

[0017] S22. Remove redundant features to improve the training efficiency of the model.

[0018] Preferably, the standardization formula for the data is:

[0019] ,

[0020] in, These are the standardized eigenvalues. These are the original eigenvalues. The mean, The standard deviation is denoted as .

[0021] Preferably, in step S3, during PCA dimensionality reduction, the top principal components with a cumulative variance contribution rate greater than 85% are selected and used as input features of the LSTM model, thereby significantly reducing data dimensionality and model input complexity while preserving the core information of the original data.

[0022] Preferably, the PSO optimization in step S4 specifically includes:

[0023] S41. Initialize the velocity and position of the particles. The position of the particles represents the combination of hyperparameters of the LSTM model.

[0024] S42. Train the LSTM model using the current particle's position (i.e., a set of hyperparameters) and calculate its prediction error (such as root mean square error), which is used as the particle's fitness value.

[0025] S43. Based on the individual optimal position and the global optimal position, iteratively update the velocity and position of all particles to guide the particle swarm to gradually approach the optimal hyperparameter combination.

[0026] Preferably, the formulas for the velocity and position of the PSO particles are as follows:

[0027] ,

[0028] ,

[0029] in, This represents the velocity of the i-th particle in the k-th iteration. This represents the position of the i-th particle in the k-th iteration. It is the optimal position for an individual. It is the globally optimal position. It is inertial weight. and It is a learning factor. and It is a random number between [0, 1].

[0030] Preferably, the core hyperparameters optimized during the particle swarm optimization process include the number of LSTM units, learning rate, training batch size, and number of training epochs.

[0031] Preferably, in step S6, when testing the model, the predictive performance of the model is comprehensively evaluated by calculating the model's RMSE, MAE, and R².

[0032] Preferably, the hybrid model exhibits superior prediction performance compared to the traditional LSTM model in experiments, specifically with a 13.04% reduction in RMSE, an 11.24% reduction in MAE, and a 0.58% increase in R².

[0033] Preferably, the final training in step S7 involves using the optimal hyperparameter combination obtained through the PSO algorithm to fully train the LSTM model, in order to ensure that the final model with the best prediction accuracy is obtained.

[0034] The beneficial effects of the present invention through the above technical solutions are as follows:

[0035] (1) The prediction accuracy is significantly improved. The core features affecting soil moisture change are effectively extracted by PCA dimensionality reduction, and redundant information interference is eliminated. At the same time, the PSO algorithm is used to automate the global hyperparameters of LSTM, avoiding the blindness of manual parameter tuning, enabling the model to find a better parameter configuration, thereby significantly reducing the prediction error and improving R².

[0036] (2) Enhanced model efficiency and generalization ability. PCA dimensionality reduction reduces the input dimension of the model, lowers the computational burden, and speeds up the training process. The introduction of PSO enables the model to adaptively find the optimal hyperparameter set suitable for different data features, thereby enhancing the model's generalization ability and robustness under different regions and soil conditions.

[0037] (3) Provide reliable decision support for precision irrigation. This method can provide more accurate and forward-looking soil moisture prediction information, which helps farmers or irrigation systems to formulate more scientific and water-saving irrigation plans, improve water resource utilization efficiency, and has important practical value for promoting the development of smart agriculture and water-saving agriculture. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0039] Figure 1 This is a flowchart of the overall workflow of the PPLSTM hybrid prediction model proposed in this invention.

[0040] Figure 2 This is a specific embodiment of the present invention, showing the cumulative variance contribution rate curve of principal components obtained by performing PCA analysis on the standardized input features.

[0041] Figure 3 This is a convergence curve of the fitness value changing with the number of iterations during the optimization of LSTM hyperparameters by the PSO algorithm in a specific embodiment of the present invention.

[0042] Figure 4 This is a comparison chart of soil moisture prediction results of the basic LSTM model and the PPLSTM model of the present invention on a test set for a certain time period in a specific embodiment of the present invention. Detailed Implementation

[0043] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments, which are preferred embodiments of the present invention. It should be understood that the described embodiments are merely some embodiments of the present invention, and not all embodiments; it should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0045] like Figure 1 As shown, a method for predicting wheat soil moisture based on a hybrid PPLSTM model includes the following steps:

[0046] S1. Obtain meteorological and soil moisture data during the wheat planting cycle and perform preliminary data screening;

[0047] S2. Perform data preprocessing on the acquired soil moisture data;

[0048] S3. Use Principal Component Analysis (PCA) to reduce the dimensionality of high-dimensional features;

[0049] S4. Apply particle swarm optimization (PSO) to optimize the hyperparameters of the Long Short-Term Memory (LSTM) network.

[0050] S5. Input the optimized LSTM network into the training data to train it and obtain the prediction model;

[0051] S6. Test the trained model using the validation set and output the prediction results;

[0052] S7. Optimize the parameters of the LSTM model based on the test results, and perform final training to obtain the model with the best prediction accuracy.

[0053] The specific steps of data preprocessing in step S2 are as follows:

[0054] S21. Fill in or delete missing values ​​in the data;

[0055] S22. Standardize all environmental characteristic variables, specifically using the Z-score standardization method, so that the mean of each characteristic is 0 and the standard deviation is 1, thus eliminating the influence of different units.

[0056] Specifically, this embodiment selects seven key environmental features: 2-meter dew point temperature, corrected precipitation, 10-meter wind direction, surface temperature, 2-meter air temperature, shallow soil moisture, and surface soil moisture. The prediction target is shallow soil moisture.

[0057] The standardization formula for the data is:

[0058] ,

[0059] in, These are the standardized eigenvalues. These are the original eigenvalues. The mean, The standard deviation is denoted as .

[0060] In step S3, the specific steps of PCA dimensionality reduction are as follows: calculate the covariance matrix of the standardized 7-dimensional feature matrix and perform eigenvalue decomposition. For example... Figure 2 As shown, the variance contribution rate of each principal component is calculated based on the eigenvalues. In this embodiment, the cumulative variance contribution rate of the first three principal components reaches 88.96%, exceeding the preset threshold of 85%. Therefore, the first three principal components are selected as the new input features of the LSTM model, successfully reducing the feature dimension from 7 dimensions to 3 dimensions.

[0061] In step S4, the specific steps for PSO to optimize the LSTM hyperparameters are as follows:

[0062] S41. Initialize PSO parameters and particle swarm: Set the number of particles to 10 and the maximum number of iterations to 15. The position of each particle represents a set of LSTM hyperparameter combinations, including learning rate, number of LSTM units, batch size, etc., and is randomly initialized within a preset range.

[0063] S42. Define the fitness function: For each particle (i.e., each set of hyperparameters), configure the LSTM model with it, train it using the training set after PCA dimensionality reduction, and calculate the predicted MSE on the validation set. Use the MSE as the fitness value of the particle.

[0064] S43. Iterative Update and Optimization: Based on the velocity and position update formulas of PSO, guide the particle swarm to search for regions with better fitness, and record the optimal positions of individual particles and the global optimum. For example... Figure 3 As shown, the MSE continues to decrease and converges as the iteration proceeds.

[0065] S44. Output Optimal Parameters: After the iteration, output the global optimal position, i.e., the optimal hyperparameter combination. In this embodiment, the optimal set of parameters obtained is: 256 LSTM units and a learning rate of 0.01.

[0066] The formulas for the velocity and position of the PSO particles are as follows:

[0067] ,

[0068] ,

[0069] in, This represents the velocity of the i-th particle in the k-th iteration. This represents the position of the i-th particle in the k-th iteration. It is the optimal position for an individual. It is the globally optimal position. It is inertial weight. and It is a learning factor. and It is a random number between [0, 1].

[0070] In step S5, the LSTM model is configured using the above-mentioned optimal hyperparameters, and the 3-dimensional principal component features obtained by PCA dimensionality reduction are used as input to train the model, thus obtaining the final PPLSTM model.

[0071] In step S6, the model performance is evaluated using an independent test set. The prediction results of the PPLSTM model are compared with those of the LSTM model using the original features and default parameters.

[0072] The quantitative performance comparison of the models is shown in the table below:

[0073] Evaluation indicators LSTM model PPLSTM model Model performance improvement RMSE 0.1510 0.1313 13.04% MAE 0.0820 0.0728 11.24% R² 0.9766 0.9824 0.58%

[0074] From the above table and Figure 4 As can be seen, the PPLSTM model proposed in this invention significantly outperforms the basic LSTM model in the three core metrics of RMSE, MAE, and R², proving the effectiveness of the PCA dimensionality reduction and PSO optimization strategy.

[0075] In summary, this invention focuses on improving the performance of LSTM models in soil moisture prediction. By introducing PCA for feature dimensionality reduction and PSO for hyperparameter optimization, a highly efficient PPLSTM hybrid prediction model is constructed. Experimental results show that this method can effectively improve prediction accuracy and provides a reliable technical solution for intelligent water management in wheat production.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting wheat soil moisture based on a hybrid PPLSTM model, characterized in that, Includes the following steps: S1. Obtain soil moisture data during the wheat growth cycle and perform preliminary screening of data characteristics; S2. Perform data preprocessing on the acquired soil moisture data; S3. Use Principal Component Analysis (PCA) to reduce the dimensionality of high-dimensional features; S4. Apply particle swarm optimization (PSO) to optimize the hyperparameters of the Long Short-Term Memory (LSTM) network; S5. Input the optimized LSTM network into the training data to train it and obtain the prediction model; S6. Test the trained model using the validation set and output the prediction results; S7. Optimize the parameters of the LSTM model based on the test results, and perform final training to obtain the model with the best prediction accuracy.

2. The wheat soil moisture prediction method based on a hybrid PPLSTM model according to claim 1, characterized in that, The data preprocessing in step S2 involves standardizing each feature of the input data according to the following formula, so that its mean is 0 and its standard deviation is 1: ， in, These are the standardized eigenvalues. These are the original eigenvalues. The mean, The standard deviation is denoted as .

3. The wheat soil moisture prediction method based on a hybrid PPLSTM model according to claim 1, characterized in that, In step S3, when performing dimensionality reduction on high-dimensional features using PCA, the top three principal components with a cumulative variance contribution rate greater than 85% are selected and used as input features for the LSTM model.

4. The method for predicting wheat soil moisture based on a hybrid PPLSTM model according to claim 1, characterized in that, The application of particle swarm optimization in step S4 specifically includes the following steps: S41. Initialize the velocity and position of the particles. The position of the particles represents the combination of hyperparameters of the LSTM model. S42. Train the LSTM model using optimized hyperparameters and calculate its prediction error as the fitness value of the particles. S43. Update the particle's velocity and position to gradually approach the optimal solution.

5. The wheat soil moisture prediction method based on a hybrid PPLSTM model according to claim 4, characterized in that, In the particle swarm optimization process in S42, the optimized hyperparameters include the number of LSTM units, learning rate, training batch size, and number of training cycles.

6. The wheat soil moisture prediction method based on a hybrid PPLSTM model according to claim 4, characterized in that, The formulas for updating particle velocity and position in S43 are as follows: ， ， in, This represents the velocity of the i-th particle in the k-th iteration. This represents the position of the i-th particle in the k-th iteration. It is the optimal position for an individual. It is the globally optimal position. It is inertial weight. and It is a learning factor. and It is a random number between [0, 1].

7. The wheat soil moisture prediction method based on a hybrid PPLSTM model according to claim 1, characterized in that, When outputting the prediction results in step S6, the predictive performance of the model is evaluated based on the calculated root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²).

8. The method for predicting wheat soil moisture based on a hybrid PPLSTM model according to claim 1, characterized in that, The final training in step S7 uses optimized hyperparameters to ensure optimal prediction accuracy.

Images