Research And Development And Irrigation Application Of Soil Moisture Prediction Platform In Lingnan Hilly Area Based On The Internet Of Things

With the growth of population and the prosperity of social development,the realization of automation,unmanned,intelligent production has become the most important to improve agricultural production efficiency.In view of the shortcomings of soil moisture measurement in China,such as high cost,poor timeliness and large error,etc.In this paper,the establishment and application of the soil moisture prediction platform based on the Internet of things in lingnan Hilly region were studied,and the meteorological data network was collected,transmitted and queried.Finally,the function of soil moisture prediction and the establishment of machine learning model in the two dimensions of dryness depth time domain are realized,which is of great significance for the efficient acquisition and utilization of soil moisture information in the research region or a large range.The main research contents are as follows:(1)According to the characteristics that affect the significant change of soil water,the variables that have a great correlation with the change of soil water were obtained by combining correlation analysis and Pearson analysis,and the variable with a small meteorological correlation,air pressure,was eliminated initially.Able to get through the analysis of the characteristics of relevance and influence to soil moisture change of characteristic variables for sunshine,air temperature,relative humidity,precipitation,wind,wind speed,wind direction,time,and different soil depth and different soil depth of maximum of absolute value of correlation coefficient of 0.93,were positively correlated,and the weaker the correlation of soil moisture,air pressure is 0.045.After the correlation analysis,the dimension of the model could not only be reduced in the subsequent modeling,but also soil moisture could be predicted more accurately after removing the variables with little correlation and retaining the significant correlation variables.(2)The sample set of soil moisture prediction was constructed and divided,simulation experiment was designed,and the correlation between each factor and the experimental results was calculated by using the feature screening tool of XGBoost algorithm.In order to achieve better prediction effect,the number of data increased from 1440 to 7200,the two weakly correlated factors of air pressure and wind direction were removed,and the main influencing factors were extracted for modeling again.The prediction accuracy of soil moisture increased from 20-80 to 95-100 zones,which further improved the prediction quality.ACC,MAE,MAPE and RMSE are also used to compare and analyze the machine learning model constructed in this paper.The performance of XGBoost algorithm is 98.89,0.28,19.82% and 0.34,respectively.The results show that XGBoost has obvious advantages in model accuracy compared with GBDT,RF,linear regression and Lightgbm algorithm,which can give full play to soil moisture prediction ability in the big data environment and provide support for the implementation of corresponding irrigation decisions.(3)Set up a soil moisture prediction platform and put forward the design scheme of each module and function.According to the characteristic that soil moisture of pakchoi growing period should be kept between 40%RH and 50%RH,matlab software fuzzy controller is designed,and matlab is connected by FLask to trigger matlab prediction algorithm to predict crop water demand and irrigation duration,and the predicted value is stored in the database system.The function combination of fuzzy control irrigation system and soil moisture prediction platform is realized.And through the Internet of things intelligent irrigation track car modeling,through MATLAB simulation analysis fuzzy reasoning calculation of irrigation time to achieve intelligent irrigation function.In the period from February 15,2022 to February 20,2022,soil moisture data with a depth of 5-10 cm were randomly sampled and compared with the actual values.The error was less than 1%,which indicated that the platform could accurately predict soil moisture,and verified the platform’s soil moisture prediction function in the time domain of dry depth-time.