Research And Application Of Machine Learning In The Decision Making System In Agricultural Internet Of Things

The data collected by wireless sensor network(WSN) in the Agricultural Internet of things system were analyzed by Hilbert-Huang Transform(HHT) in this thesis. The data collected by wireless sensor networks has the characteristics of nonlinear and non-stationary. And HHT is able to completely adaptively decompose the data with nonlinear, non-stationary characteristics into multiple intrinsic mode functions(IMF) and a residue. Thus, it provides a new insight for data analysis in time and space. The method has been verified in the WSN node data collected in Yanqing, Beijing.Then, machine learning method is introduced into the decision making system. Based on the HHT decomposition, a number of intrinsic mode functions(IMF) and residue are analyzed and modeled. So it has formed a set of system of data analysis, decision-making method. The method has been validated in the samples of data of precipitation in Yunnan area.Specifically, the main contributions of this thesis are as follows:(1) The theory and application of HHT, linear regression, artificial neural network(ANN) and support vector regression(SVR) are studied in depth, which enrich the research method of data analysis in data analysis subsystem.(2) The HHT data processing method is applied to the processing of the data collected by wireless sensor networks in the data analysis subsystem. The data of crop growth environment variables collected from 36 sets of WSN nodes(covering area close to 100 km2) in farmland of Yanqing District of Beijing city are analyzed. The data including soil moisture, soil surface temperature, photons, which are important variables in ecosystem models, totally more than 230000 samples of data. Based on the nonlinear and non-stationary characteristics, HHT method is used to decompose it into a number of IMF and a residual term. From different scales of IMF components to analysis inherent characteristics and physical meaning of the original data. HHT data processing method is used to analyze the temporal correlation of different types of data and the correlation of the same type of data based on the location of the different geographic. It is prepared for the Agricultural Internet of things to make decision.(3) In the intelligent irrigation system, HHT data processing method is first used to deal with the real precipitation samples of data in four regions of Yunnan(EEMD decomposition). Then the machine learning method is applied to establish the precipitation forecast model. Several models including artificial neural network, support vector regression and multiple linear regression models are investigated in this thesis. Experimental results show that, EEMD combined with machine learning method showed a fairly accurate prediction effect and the performance of the model is very stable. Then put forward an innovative mixed model.In summary, two subsystems in the Agricultural Internet of things decision making system have been studied: data analysis subsystem and intelligent decision subsystem. In the data analysis subsystem, the time-spatial correlation analysis of the HHT based farmland environment variable samples of data is analyzed in different scales. The application of machine learning in decision making system is studied in the intelligent irrigation system.