Research On The Application Of Bayesian Network In Agriculture Intelligent System

There exists a lot of uncertainty phenomenon and problem. Uncertainty in agriculture is more extensive and complex. So, in order to create an effective intelligent system, uncertain knowledge must be dealt with. From representation of uncertainty knowledge, there are two methods of dealing with uncertainty. One is rule-based method, the other is model-based method. The advantage of rule-based method is that its computation is convenient, its disadvantage is that its syntax is not systemic. The advantage and disadvantage of model-based method is contrary to the rule-based method. From measurement of uncertainty, the methods of dealing with uncertainty are fuzzy theory and probability theory. Fuzzy theory mainly deals with vagueness, and probability theory mainly deals with randomness. Therefore, Bayesian network in the article is a model-based probability method. Bayesian network is a combination of probability theory and graph theory. Study of Bayesian network originate from the 1980's.Since 1990's,its study and application has stirred great concern .Compared with rule based method, the syntax of Bayesian network is more clear, it can reason in dual direction and can be constructed and debug rapidly. The disadvantage of Bayesian network is that the compute complexity is high. IT, especially artificial intelligence application in agriculture is the key to agriculture modernization. Construction of intelligent system is of great importance for improving production rate and using production knowledge. This article mainly introduces Bayesian network usage in agriculture. There are two problems to be dealt with. One is the construction of Bayesian network model, the other is the implementation of inference algorithm. When Bayesian network is to be created, it needs to get the Bayesian network structure and CPT. there are three methods to get the two parameters. The first is auto achieving by machine learning, the second is manual achieving by communicating with field expert, the third is half auto achieving by combination of field expert and machine learning. As far as china agriculture concerned, it is difficult to achieve much data. So, it is not fit to create Bayesian network. This article introduces the manual construction method. Bayesian network inference algorithm can be classified with exact algorithm and approximation algorithm. But people have proved that both of them are NP-hard. Exact algorithm is fit for small network, and approximation algorithm is fit for large network. This article introduces variable elimination in exact algorithm and sampling algorithm in approximation algorithm, expert system developer can use different algorithm according to different situation. Manual method needs to get network structure and CPT. It is a process that is from abstract to concrete and from quality to quantity. Structure should be conformed before CPT .structure should be as simplified as possible. This can be achieved by divorcing parent nodes. By doing this, the number of CPT can be reduced largely, then the problem complexity can be low. The CPT achieving is a very difficult problem that all people acknowledged. In this article, we introduce noisy-or mechanism that can get CPT by less probabilities. The idea of VE is: the variable set in the network is divided into three parts, namely evidence nodes, query nodes and hidden nodes. When compute posterior probability, sum-product algorithm was used which compute summation for hidden nodes and product every node's CPT. The idea of sampling algorithm is: according to topology sequence, number from random generator is compared with node's prior probability, then a bet ring is formed to choose nodes state and assign value to it. When all nodes are assigned once, a network sample was achieved. Repeat it until a sample (m samples) was achieved. At last, the probability can be got. In general, an intelligent system is of decision ability. Decision theory is combination probability theory and utility theory, after decision nodes and utility nodes was added, Bayesian network become decision network which enhance intelligent system. After the theory about Bayesian network construction and inference was discussed, Bayesian construction component and inference component which was developed with c# language. The function of construction is to create Bayesian network and save the structure and CPT parameters in the XML file. In the inference component, we can first open the former XML file, then use...