Prediction Of Soil Moisture Based On Deep Neural Network

Soil moisture prediction is one of the important contents of agricultural development,and it is of great significance for achieving crop yield increase and income increase and improving crop economic benefits.Reasonable control of water supply and fine crop management.This paper uses deep neural network algorithm to predict soil moisture and realize the scientific supply of tea garden irrigation water.Soil moisture is a type of data with time series nonlinear properties.Currently,more applications for time series are recursive deep neural networks.LSTM deep neural network is good at exploring the nonlinear relationship between time series data,which is suitable for soil moisture sequence prediction.To this end,the main research contents and results of this paper are as follows:(1)The difficulty and the common methods for predicting soil moisture are introduced in detail.And the basic concepts of neural networks are elaborated.It includes BP neural network,cyclic neural network(RNN),long-term and short-term memory network(LSTM)basic concepts of several common neural network models,structure diagram,derivation process and so on.(2)Collecting data on soil moisture data and its influencing factors,and normalizing the data,distributing the data in the range of[0,1],and then performing correlation analysis to extract data with large impact factors on soil moisture.The input of the model.From the aspect of model construction,the models of different hidden layer neurons are trained and the training results are compared and analyzed to illustrate the influence of model structure on prediction accuracy.In the optimization method,the network training process is optimized by SGD,RMSprop and Adam methods respectively,to study the influence of different optimization methods on the accuracy of the model,and finally select the optimization method most suitable for LSTM deep neural network.The choice of time series is based on the comparison and analysis of the prediction results of different time series,and the time series with the highest prediction accuracy is the time series of LSTM deep neural network.(3)The data set used is the meteorological and soil moisture data collected in the 2018-year and third months of the Huangshan Taiping area.The sensor is continuously collected 24 hours a day,and the sampling frequency is once per minute.Through the analysis of experimental results,the prediction accuracy of deep neural network model is more accurate than that of BP neural network and multivariate quadratic regression model.The research shows that the deep neural network model has a good application prospect in soil moisture prediction,and is based on depth.The establishment and research of soil moisture prediction model based on neural network can provide a reference for soil prediction in other regions.In this paper,a method of fine soil moisture prediction based on deep neural network technology is proposed.The application of time series method in soil moisture prediction is studied,and the regional soil moisture prediction method is expanded.