The Application Of Bayesian Networks In Land Use Behavior Of Farmers Participation In Retired Farmland

Based on the complex behaviors of farmers using their land under the backgrounds of grain for green program and urbanization construction and the non-standard change of applying Bayesian network to land use/cover, this thesis proposed a standard Bayesian network modeling method. By using the BNT toolbox of Matlab? and the Fayyad's MDL discretization algorithm one determines the discrete optimum cut-off point survey data and gets three sets of data. Then, 36 networks are constructed by using 12 kinds of structure learning algorithms. The MWST+T-K2 structure learning algorithm, which has the highest Bayesian score, the highest LL score, the highest AIC score, the higher BIC score, the acceptable running time, greatly meeting with the reality and certain extension value, is selected to construct the three networks to research the behaviors of farmers using their land in returning farmland to forest(BF-LUC-BN).Two methods, CV-5 and CV-10, are used to classify the three kinds of Bayesian classifiers, and the classification accuracy, the error bars, the correct rate of classification, the recall rate and the F1- score of BF-LUC-BN are obtained. By comparing these indexes with NBC, TANC and MWST classifier, it is found that MWST+T-K2 classifier has the highest classification level on network of two states, but the classification level of that on the network of three states and the network of four states is slightly lower than the classification level of the TANC, and the more accurate the classification number, the lower the accuracy of the data. The structure of the two network is improved by using combined reasoning, and the error of the overall reasoning of three network and four network is not more than 0.015, which illustrate the effectiveness of the network. At the same time, the results show that the subsidy for returning farmland to forest should be set to be small or medium amount of subsidy.Comparing the experience Bayesian network(EI-BF-LUC-BN) of the returning farmland to forest farmers land behavior with the TANC and MWST+T-K2 classifiers classification test results, it is found that only the EI-BF-LUC-BN of the four network has advantage. At the same time, based on Monte Carlo algorithm, the improved network(MI-BF-LUC-BN) based on Monte Carlo algorithm is proposed, the network structure of the largest parent node and random number generation is discussed, and the stability of the network is studied. The results show that MI-BF-LUC-BN can identify the combinations without data.