Analysis And Prediction Of Crop Environmental Factors

In crop production,increasing the yield of crops is one of hot topics in today's research,and it is also an important research direction in crop research projects.The growth environmental factors of crops are inextricably linked to its yield,so the research on environmental factors is inevitable.This thesis mainly analyzes the changing laws of the environmental factors of crops and themselves,and at the same time studies their forecasting methods.First of all,for this thesis,the experimental data needs to be preprocessed,that is,to eliminate the abnormal redundant values in them,to get a complete and clean and reliable data set.Then,three aspects of research work are carried out on environmental factors: the first one is the correlation analysis of environmental factors,analyzing the correlation between environmental factors and their shelf-correlations based on Pearson correlation,and selecting the environmental factors with strong correlation for specific analysis;the second one is to study the distribution law of environmental factors by using the QQ diagram and S-W data distribution test method to verify whether the air temperature and illuminance conform to the normal distribution,or other specific distribution for atmospheric pressure and soil temperature etc;the third one is to fit and predict the environmental factors where some fitting functions are used to fit all environmental factors over the whole data set and daily data,and the environmental factors are divided into two categories according to the fitting results.The first type of the environmental factors is defined as poor fitting including soil moisture,soil salinity,wind speed,PM2.5,rainfall,evaporation and carbon dioxide concentration;the second type of the environmental factors is defined as good fitting including air temperature,air humidity,soil temperature,atmospheric pressure and light intensity.Two combination forecasting models according to the own laws of those two types of data are used,in which the first one is SSA-EMD(1/4)-LSTM model where we firstly use the singular spectrum analysis method SSA and empirical mode decomposition method EMD(1/4)algorithm to decompose the environmental factors twice,and then use the long and short memory network LSTM finish predictions;the second one is SSA-LSTM model where we firstly use the singular spectrum analysis method SSA to decompose environmental factors to extract the trend,periodicity and noise of the data,and then use the LSTM neural network to predict the decomposed data.Experimental results show that the first combination prediction model predicts the first type of environmental factors compared to other models(LSTM,SSA-LSTM,EMD(1/2)-LSTM,EMD(1/4)-LSTM,SSA-EMD(1/2)-LSTM)prediction mean square error and mean absolute error decreased by 0.516 and 0.193 respectively.The second combination prediction model predicts the second type of environmental factors compared to other models(LSTM,EMD(1/2)-LSTM,EMD(1/4)-LSTM,SSAEMD(1/2)-LSTM,SSA-EMD(1/2)-LSTM)prediction mean square error and mean absolute error decreased on average by 0.855 and 0.366 respectively,indicationg that our model are efficient.