Research On Soil Moisture Prediction And Irrigation Decision Of Winter Wheat Root Zone Based On Data Mining

The contradiction between supply and demand of water resources in China is becoming more and more significant,and irrigation water accounts for a large proportion in agriculture.As the scale of agricultural industrialization continues to expand,the requirements for agricultural production and water-saving irrigation are increasing.Ensuring that the soil water content is maintained at the optimal state suitable for crop growth through reasonable irrigation decisions will become an effective way to alleviate the contradiction between supply and demand of agricultural water.This paper uses deep learning algorithms to construct a daily rainfall prediction model based on actual meteorological data and website forecast rainfall data in the experimental area,and constructs a daily crop water demand prediction model for winter wheat fertility based on daily predicted temperature,and finally applies the established model to the field water balance equation to construct a soil water prediction model for winter wheat fertility,and to The model was applied to the field water balance equation,a soil water prediction model was constructed,and the irrigation system was optimized for winter wheat.The research contents and main conclusions of the paper are as follows.（1）A rainfall prediction model based on long-and short-term memory networks was established.The four meteorological factors with high correlation coefficients（minimum temperature,average temperature,dew point temperature,and average atmospheric pressure）were used as input variables of the rainfall prediction model through correlation analysis,and the LSTM algorithm was used to optimize the website forecast rainfall data,and the forecast evaluation parameters R²,RMSE and RPD were obtained as 0.875,3.120 and 2.787,respectively.website forecast rainfall data evaluation The parameters R²,RMSE and RPD were 0.822,3.719 and 2.523,respectively.the optimization of website forecast rainfall could not be achieved using CNN algorithm and SVM algorithm,indicating that the LSTM model has the best prediction performance and model accuracy,and is suitable for accurate prediction of small-scale rainfall in the field.（2）The daily minimum temperature and daily maximum temperature prediction models were constructed using measured meteorological data with 10 and 4 factors as input variables,respectively.First,the daily minimum temperature prediction models were constructed with10-factor and 4-factor as input variables,and the prediction results were 0.941 and 0.933 for R²,2.178 and 2.289 for RMSE,and 3.804 and 3.614 for RPD,respectively.The prediction results of each model were better than the forecasted temperature values on the website.（3）The Hargreaves formula was corrected for the regional parameters using the measured meteorological data for the past 11 years,and the corrected C,E,and T parameters were0.00067,0.54,and 21.4,respectively.10-factor ET₀ prediction models were constructed by substituting the daily temperatures predicted using 10 meteorological factors into the corrected Hargreaves formula,and the daily temperatures predicted using 4 meteorological The daily temperature predicted by using 4 meteorological factors was substituted into the modified Hargreaves formula to construct a 4-factor ET₀ prediction model;the website forecast temperature was substituted into the modified Hargreaves formula to construct an ET₀ website prediction model.The prediction results of the three models were 0.887,0.877 and 0.804 for R²,0.511,0.549 and 0.673 for RMSE,and 2.854,2.552 and 2.166 for RPD.The prediction results showed that the 10-factor ET₀ prediction model was the best and the 4-factor ET₀prediction model was slightly worse,and both models had excellent prediction performance and can be selected according to the abundance of meteorological data in the irrigation area,and the ET₀ site prediction model is the least effective.According to the three ET₀ prediction models,the 10-factor ET prediction model,the 4-factor ET prediction model and the ET site prediction model were established throughout the fertility period of winter wheat using the two-crop coefficient method,and the results showed that the 10-factor ET prediction model had the best effect,while the 4-factor ET prediction model could maintain high accuracy while using fewer meteorological factors and was more suitable for crop water demand prediction.The ET site prediction model built by site forecasting temperature is less accurate,but the data is easy to obtain and is suitable for use in areas where field meteorological data are lacking.In summary,the prediction model of crop water demand based on predicted temperature is effective and can be used for daily crop water demand prediction.（4）Based on the constructed daily rainfall prediction model and daily crop water requirement prediction model,a soil moisture prediction model for winter wheat fertility was constructed by applying the principle of field water balance.The average daily relative error of the prediction results was 1.69%,indicating that this model can predict the daily variation of winter wheat root water better.The soil moisture prediction model was applied to irrigation decision optimization by determining the irrigation time and irrigation water quantity with the given upper and lower limits of soil moisture content during the fertility of winter wheat.After irrigation decision optimization,the number of irrigation times increased but the amount of irrigation water decreased during the fertility of winter wheat,and the overall water saving was about 20%;in addition,the soil moisture content of the wet layer of winter wheat plan could be maintained within the appropriate range after irrigation decision optimization,which was beneficial to the growth and development of the crop.The results show that the soil moisture prediction model and irrigation decision optimization mechanism established by the data of the moisture field test area in this paper are effective and can be used in the design of intelligent irrigation system.