Research On Data Processing Method Of Decision Making System In Agricultural Internet Of Things

Agricultural Internet of Things is a specific application of networking technology in agricultural production and management. Decision system is a very important part of agricultural Internet of Things. It’s responsible for data processing, analysis, reasoning and decision-making and provides decision support to facilitate production and management of agriculture. Grain yield prediction is an important function in decision system. Grain yield prediction is beneficial for breed improvement of crops, the adjustment of grain-production structure, policymaking of local and national government. Additionally, it’s also conductive to speeding up construction and development of Chinese modernization of agriculture. It has great significance about grain or food security, national economic development and national security. As a result, scientific prediction of grain yield has an important strategic position. In this thesis, taking maize yield of Mengzi county of Yunnan province as an example, we conduct research on prediction method of crop yield in order to improve the prediction accuracy for crop yield.Main research work in this thesis contains the following fields:(1) Summarize the current researches of crop yield prediction method at home and abroad, and then choose multiple linear regression(MLR), artificial neural network(ANN) and support vector machine(SVM) as yield prediction methods, and then analyze their theory.(2) Using MLR method, ANN method and SVM method, we make prediction about the maize yield in Mengzi county Yunnan province, and analyze their prediction results and forecast error to judge whether the models meet precision requirement or not. Besides, four accuracy metrics were used to measure the overall predictive ability: the correlation coefficient(R), the mean absolute error(MAE), and the root mean square error(RMSE).(3) We make comparative study on the established MLR model, the ANN model and the SVM model from the two aspects——model test results and theory, to get the final conclusion.On the basis of previous researches, this thesis adopts the MLR method, ANN method and SVM method to predict the maize yield in Mengzi county Yunnan province, and we obtain the following conclusions:(1) The established MLR, ANN and SVM prediction model can be well applied to forecast maize yield, and each prediction error of the three models is relatively small. They all get high prediction accuracy.(2) In MLR model, the maize yield is closely relative to the average daily amount of reference crop evapotranspiration(ETO) in May, the highest temperature in August, the lowest temperature of June and the lowest temperature in August, however in the ANN and SVM model, the maize yield is also relevant with the average daily amount of reference crop ETO in March and April。(3) Real data of maize yield in Mengzi county Yunnan province is used for building and validating the models. The validating results show that MLR model has the highest R value(0.813), the lowest MAE(0.1044) and the lowest RMSE(0.1441); Followed by SVM model, R is 0.8046, MAE is 0.1179 and RMSE is 0.1546; And ANN obtains the lowest R value(0.7479), the highest MAE value(0.1326) and the highest RMSE value(0.1691). We can conclude that the MLR model is the best, followed by the SVM model, and the ANN model is the worst.