| Alfalfa is a forage crop of cattle and sheep with high nutritional value.It is mainly planted in arid and semi-arid areas and its growth depends on rain and irrigation.Therefore,a stable soil moisture prediction technology is needed to dynamically simulate the change of soil moisture,so as to achieve the purpose of scientific and rational irrigation and effective utilization of water resources.In this paper,alfalfa is taken as the research object,aiming at the inaccurate prediction of soil moisture caused by the uncertainty of rainfall and the randomness of man-made irrigation,the deep learning method is used to study and discuss this problem.The main content of this paper is as follows:1.A data set on alfalfa growth environment was constructed.Through literature review,relevant public data sets are lacking.Therefore,to plant alfalfa in Ningxia agricultural reclamation experimental land,design the experimental scheme in the early stage,set up soil moisture sensors and meteorological monitoring stations in the experimental land,collect data regularly,sort out historical data,and construct a meteorological data value from April to September in 2017 and 2018 and from July to October in 2022,artificial irrigation value,The data set composed of soil moisture values of different soil depths contains 19,763 pieces of data,which provides data support for the subsequent prediction of soil moisture of alfalfa multilayer.2.Through comparative analysis of multi-layer soil water prediction of alfalfa under different single-branch models,it was found that rainfall or irrigation would cause inaccurate model prediction of soil water.In order to compare and analyze the applicability and accuracy of different single branch models in soil water prediction at multi-layer depth of alfalfa,this paper established artificial neural network models,long short-term memory network models and bidirectional long short-term memory network models that can be used to predict the soil water content at multi-layer depth of alfalfa at different growth stages.The experimental results showed that the artificial neural network model performed well,while the bidirectional long and short term memory network model had poor prediction effect.The prediction effect of the three models was better at the depth of 30cm soil layer in the branch bud stage,but worse at the depth of 20cm soil layer in the regeneration stage and 10cm soil layer in the first flower harvest stage,showing a large error.3.Aiming at the prediction inaccuracy caused by the uncertainty of rainfall and the randomness of man-made irrigation,this paper proposes a prediction method based on a two-branch model.The model has a two-branch structure.The right branch is composed of a convolution based residual network to model the dependence relationship between soil water,and the left branch is composed of a fully connected layer to map rainfall and irrigation characteristics.The output results of the two branches are fused to improve the accuracy of predicting soil water values at multiple layers in different growth stages of alfalfa.The experimental results showed that the root-mean-square error of the predicted and predicted soil water content of alfalfa was less than 1.23.The prediction ability of the two-branch model under different delay days is analyzed and compared.The prediction effect of the model is stable under different delay days.The prediction results of different single branch models are analyzed and compared,and the prediction results of the model proposed in this paper are better than the other three models.4.The construction of a monitoring and prediction system for alfalfa soil moisture.The system uses the open-source framework Streamlit and Python technology,combined with deep learning,to achieve data visualization,data preprocessing,model selection,and result display functions.It can monitor and predict soil moisture during the alfalfa growth process and has practical significance for guiding scientific irrigation. |
PDF Full Text Request