| Precision agriculture is the effective shortcut to implement Agricultural Sustainable development. Getting accurate and reasonable decision by dealing with agricultural data and knowledge is the core content of precision agriculture. With the development of agriculture and the progress of technology, some new cases and problems appeared. Traditional methods of treating data can't meet the need of precision, digitization and intellectualization of agricultural production. In this thesis, guided by agricultural systematology and taken maize precision production as research object, according to the problem existing in construction of fertilization model,evaluation of land capability, yield forecasting, division of management zones and developing of agricultural intelligent decision system, based on Bayesian network, fuzzy clustering, neural network ensemble, expert system, etc, some new methods for precision agriculture decision are proposed, which result in good effect.1 . According to the complexity and uncertainty of precision agriculture information the agricultural system theory is introduced. Then the intelligent decision technology architecture is constructed.(1)Guided by agricultural system theory, the requirement of precision agriculture decision is analyzed. The solving of precision agriculture problem is divided into three layers, namely data, knowledge and decision.(2)It is pointed that spatio-temperal quality and uncertainty is the potential nature of precision agriculture information. Then the precision agriculture technology architecture with the shape of hourglass is constructed.(3)The method of combining intelligent decision technology with precision agriculture is given. Finally,Bayesian network,neural network, etc used in this paper are introduced.2.A novel neural network ensemble method is proposed and used to construct precision fertilization model.(1)A neural network ensemble method based on AP clustering and Lagrange Multiplier is proposed. In which, the method of sampling with replacement is used to produce networks and a novel formula is defined to measure the similarity of networks. By the AP clustering algorithm,networks with high precision and greater diversity are selected. Then by the Lagrange Multiplier(LM) method, these networks are combined. The experiment on the standard dataset shows that, the algorithm has higher accuracy and stronger generalization than the single neural network. Furthermore, as a linear weighted ensemble method, LM ensemble is better than Forecasting Effective Measure(FEM)ensemble and equal weight.(2) Base on the above research,two kinds of fertilization model based on neural network are proposed. One is the 4-x-3 model,the other is 6-x-1 model. The result shows that both models can simulate the nonlinear relation existing in soil nutrient, fertilization rate and yield. Especially,the 6-x-1 model not only can compute fertilization rate precisely but also can forecast yield. The practice shows that neural network ensemble based fertilization model is better than traditional fertilization models and existing neural network based fertilization models.3.Bayesian network is a strong tool of dealing with uncertainty knowledge, which has the advantage of bi-direction reasoning, fusion of prior knowledge, etc. In this thesis, Bayesian network is used in precision agriculture decision.(1)A method of manual constructing Bayesian network through integrating expert experience and"noisy sum"model is proposed. Then the method is used to construct decision network for precision agriculture, which can make decisions by different utility.(2)A Bayesian network for yield forecasting is constructed. Independence analysis shows that when the N,P ,K is known,the yield is determined without knowing H2O and organic matter. Sensitivity analysis shows that plant height in crop trait and P in soil are very important factors impacting yield.(3)Land classification model based on Bayesian classifier is constructed. The results show that both TAN model and na?ve Bayesian model have higher accuracy rate(larger than 82%),so they can be adopted in actual production. On the other hand, the accuracy rate of TAN model is a little larger than na?ve Bayesian model, which verifies that there exists correlation to some extent among the attributes affecting land capability. Finally, by sensitivity analysis for the na?ve Bayesian model, it is concluded that salinity, rainfall and terrain are important factors affecting land capability.4.Three methods of dividing management zones are proposed.(1)A division method based on directed variation degree is proposed. The results show that the directed variation degree method is accordance with the practice that the different variation degree exists in the different direction in the field.(2)A division method based on grey association degree is proposed. In this method, yield data under different treatments in multi points is analyzed by grey association degree, then the management zones is divided by the interval of grey association degree. The results show that this method is suitable for the fertilizer effect data and the situation of scare data.(3)A division method based on weighted fuzzy clustering is proposed. In this algorithm, the determination of the weight of different soil nutrients is a key step. Two methods are used to determine the weight, one is the primary component analysis(PCA), the other is coefficient of variation (C.V).To verify above methods, at the experimental site, a field in Yushu city in Jilin province, N, P, K were sampled, then the weighted fuzzy clustering algorithm was used and the number of the management zones is determined. Then the effectiveness of the weighted fuzzy clustering method is validated by comparing with the traditional fuzzy clustering method. The results shows more practical and can be used to direct precision fertilization.5.Aiming to the precision, intelligence and networking of maize production,a intelligent decision system for maize precision production based on web is developed. The features of the system are as follows::(1)The system integrates GIS and ES,which makes it deal with spatial information and uncertainty information.(2)The system integrates two kinds of precision working mode. One is based on GPS grid,and the other is based on farmer field.The practice results show that the application of the system increases the yield, decreases the price and lesson the pollution.
|
PDF Full Text Request