CN-121997038-A - Wheat soil moisture prediction method based on mixing PPLSTM model

CN121997038ACN 121997038 ACN121997038 ACN 121997038ACN-121997038-A

Abstract

The invention discloses a wheat soil moisture prediction method based on a hybrid PPLSTM model, which comprises the following steps of S1, obtaining soil moisture data in a wheat growth period, carrying out preliminary screening on data characteristics, S2, carrying out data preprocessing on the obtained soil moisture data, S3, carrying out dimension reduction processing on high-dimensional characteristics by using Principal Component Analysis (PCA), S4, optimizing super parameters of a long-short-period memory (LSTM) network by using Particle Swarm Optimization (PSO), S5, inputting training data into the optimized LSTM network to obtain a prediction model, S6, testing the trained model by using a verification set, outputting a prediction result, S7, optimizing parameters of the LSTM model according to the test result, and finally training to obtain a model with optimal prediction precision. The invention solves the problems of redundancy of high-dimensional input features, insufficient capturing capacity of time dependence, dependence on experience of super parameter selection and the like in the traditional method by combining a PSO optimization algorithm, a PCA dimension reduction algorithm and an LSTM network. Compared with the 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 water-saving irrigation practice is effectively promoted, and the sustainable development of accurate agriculture is promoted.

Inventors

SONG CAIXIA
WANG TING
CHEN LIJIAN
ZHAO YUNLONG
GAO JIWEI
SONG XINQUAN
XIE JINBAO
WANG QISHUO
WEI GUANGSEN
WANG JIANLIN
Han Taoyang
CHU JIANXIANG

Assignees

青岛农业大学

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (8)

1. A wheat soil moisture prediction method based on a mixed PPLSTM model is characterized by comprising the following steps: S1, acquiring soil moisture data in a wheat growth period, and primarily screening data characteristics; s2, performing data preprocessing on the obtained soil moisture data; S3, performing dimension reduction processing on the high-dimensional features by using Principal Component Analysis (PCA); S4, optimizing super parameters of a long-term memory (LSTM) network by using Particle Swarm Optimization (PSO); S5, inputting training data into the optimized LSTM network 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 the test result, and finally training to obtain the model with optimal prediction precision.
2. The wheat soil moisture prediction method based on the hybrid PPLSTM model according to claim 1, wherein the data preprocessing in step S2 is to normalize each feature of the input data according to the following formula, so that the mean value is 0, and the standard deviation is 1: , Wherein, the As a result of the normalized feature values, As the original characteristic value of the object is obtained, Is the mean value of the two values, Is the standard deviation.
3. The wheat soil moisture prediction method based on the hybrid PPLSTM model according to claim 1, wherein the PCA in step S3 selects the first three principal components with a cumulative variance contribution rate greater than 85% as the input features of the LSTM model when performing the dimension reduction processing on the high-dimensional features.
4. The wheat soil moisture prediction method based on the hybrid PPLSTM model according to claim 1, wherein the application particle swarm optimization in the step S4 specifically includes the following steps: S41, initializing the speed and the position of the particles, wherein the position of the particles represents the super-parameter combination of the LSTM model; S42, training an LSTM model by using the optimized super parameters and calculating a prediction error of the LSTM model to serve as an adaptability value of the particles; s43, updating the speed and the position of the particles, and gradually approaching to the optimal solution.
5. The method for predicting the moisture content of wheat soil based on the hybrid PPLSTM model as set forth in claim 4, wherein the super-parameters for optimization in the particle swarm optimization in S42 include LSTM unit number, learning rate, training lot size, and training cycle number.
6. The method for predicting the moisture content of wheat soil based on the hybrid PPLSTM model as set forth in claim 4, wherein the updated particle velocity and position formula in S43: , , Wherein, the Representing the velocity of the ith particle in the kth iteration, Indicating the position of the ith particle in the kth iteration, Is the optimal position of the individual and, Is the global optimum position for the device, Is the weight of the inertia, which is the weight of the inertia, And Is a factor of the learning process, And Is a random number between 0, 1.
7. The method according to claim 1, wherein the prediction performance of the model is evaluated based on a Root Mean Square Error (RMSE), a Mean Absolute Error (MAE) and a decision coefficient (R2) of the calculation model when the prediction result is outputted in the step S6.
8. The wheat soil moisture prediction method based on the hybrid PPLSTM model according to claim 1, wherein the final training in step S7 uses optimized super parameters to ensure optimal prediction accuracy.

Description

Wheat soil moisture prediction method based on mixing PPLSTM model Technical Field The invention relates to the technical field of wheat soil moisture prediction, in particular to a wheat soil moisture prediction method based on a mixed PPLSTM model. Background In wheat production management, accurate regulation and control of soil moisture has a key effect on crop growth safety, and the accurate regulation and control of soil moisture not only directly relates to the yield of wheat, but also can influence the moisture utilization efficiency, and further relates to the stability of grain production and sustainable utilization of agricultural resources. The development of deep learning technology brings new opportunities for soil moisture prediction, wherein a prediction model based on LSTM time sequence data becomes an important technical support for realizing intelligent irrigation decision-making by virtue of the efficient processing capacity of the prediction model. However, building an efficient soil moisture prediction model in an actual farmland environment still faces a series of serious challenges. When the traditional prediction method is used for processing multi-source and high-dimensional soil and environmental characteristics, the input variables are redundant and associated with complexity, so that the model training efficiency is low, the fitting problem is easy to occur, and key factors which really influence the moisture change are difficult to extract. Meanwhile, although the prediction models such as LSTM can effectively capture the time sequence dependency relationship, the super-parameter setting has obvious influence on the prediction performance, the traditional manual parameter adjustment mode is time-consuming and labor-consuming, and the optimal balance between the convergence speed and the prediction precision is difficult to realize, so that the reliability and the practicability of the model in real-time irrigation decision are limited. Patent document CN119357606a discloses a soil humidity prediction method based on an external climate environment, which processes historical data using a gray model and generates predicted values at various future time points, and then corrects the predicted values using an LSTM model and outputs final predicted values. However, the method has high input characteristic dimension and intrinsic correlation, and the original data is directly input into the LSTM model, so that the calculation complexity of the model is increased, noise is more easily introduced, and the model is difficult to focus on the most critical climate and soil humidity change mode, thereby influencing the robustness and efficiency of prediction. The patent document CN118940160A discloses a soil humidity prediction method based on an attention coding decoding LSTM model of a physical process, which utilizes an HBV hydrologic physical model to generate intermediate variables with physical significance from original meteorological and land data, screens key features through correlation analysis, inputs the physical features and the original data into an integrated attention mechanism coding decoding LSTM model for training, and finally utilizes the model to realize the prediction of future surface soil humidity. However, the method uses the coding and decoding LSTM model of the attention mechanism, has huge parameter space, uses fixed learning rate, initialization weight and other settings for training, is very easy to sink into a local optimal solution, has low convergence speed, has unstable training process, increases the overall calculation cost of the system, and reduces the prediction precision. The present invention provides a new solution to this problem. Disclosure of Invention Aiming at the situation, in order to overcome the defects of the prior art, the invention aims to provide a wheat soil moisture prediction method based on a total PPLSTM model so as to solve the problems existing in the prior art, and the specific scheme is as follows: a wheat soil moisture prediction method based on a composite PPLSTM model comprises the following steps: S1, acquiring soil moisture data in a wheat growth period, and primarily screening data characteristics; s2, performing data preprocessing on the obtained soil moisture data; S3, performing dimension reduction processing on the high-dimensional features by using Principal Component Analysis (PCA); S4, optimizing super parameters of a long-term memory (LSTM) network by using Particle Swarm Optimization (PSO); S5, inputting training data into the optimized LSTM network 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 the test result, and finally training to obtain the model with optimal prediction precision. Preferably, the data preprocessing in step S2 includes normalization proc