Comparison Of Fuzzy Evaluation And BP Neural Network Applied For Cultivated Land Quality Evaluation Of Fengtai County

The monitoring and evaluation of soil fertility was carried out by the Ministry of Agriculture in the country since 1984.Fertility evaluation was a comprehensive reflection of the quality of Cultivated land.Improving farmland productivity evaluation precision to make them more practical value and to guide agricultural production.It made perfect sense.In this paper,a case study at Huaihe Plain Fengtai County,thematic maps and text data which had been collected were vectored.Then all the vector maps were overlaid to form of arable land.Remove the outliers in text data,sampling sites were produced in ArcGIS,than the data was put in the Arable unit by interpolation.The arable unit containing property was evaluated by different evaluation models,what were traditional method(Fuzzy AHP evaluation method)and BP neural network model,and comparison of the accuracy of the evaluation results.First of all,arable unit was evaluated by traditional methods(Fuzzy AHP evaluation method).The evaluation factors were selected which were closely related Fcultivated land of Fengtai County.The evaluation factors included Soil parent material,topography,topsoil texture,topsoil bulk density,CEC,pH,organic matter,phosphorus,potassium,available Zn,Fe and Mn,a cross-sectional configuration,topsoil thickness,irrigation,and drainage conditions.There were 16 factors in all.Hierarchical model was established,so that the weight of each factor was calculated.Membership functions were fitted on the basis of each index,so membership was owned by each index.Rating was calculated for each unit of arable land by accumulating method.All the arable land was divided into five levels based on national standards.The first and second grade arable land,which was counted out,accounted for 90.2%.Secondly,Fengtai County arable land had been evaluated by the BP neural network model.The input data was the 16 evaluation factors,and the output data was the annual target yield.The arable land of Zhangji town,Guqiao town,Guiji town,north of the town,xinji town,Liuji township,Economic Development and Lee Chong township were the input data,and the arable land of Daxing town,Guandian township DingJixiang township were the validation data,and the arable land of Shangtang town,Zhuma town,Yangcun township,Gudian township,Qianmiao township were forecast data.In the BP neural network model,tansig was chosen as the input function,and purelin was chosen as the transfer function,and trainlm was chosen as the training function.There were 14 nodes inthe hidden layer,the training accuracy was over 0.93.The farmland productivity of Shangtang town,Zhuma town,Yangcun township,Gudian township,Qianmiao township was predicted by BP neural network model which had been built.The results was divided according to the national standard of cultivated land fertility again.The proportion of each level fertility was calculated,of which the first and second arable land was 91.3%.Finally,the evaluation result that was evaluated by different models was compared with the real farmland productivity.Each level fertility were counted by town,and the spatial distribution of different levels cultivated land fertility was drawn.Final results showed that the relative error of the first and second grades of cultivated land was 0.04 and 0.06,less than 0.20 and 0.11 of the conventional method.Spatial distribution of the result,which were evaluated by BP neural network evaluation,was closer to the true value than traditional method(Fuzzy AHP evaluation method).The result of Fengtai County fertility evaluation showed that Neural network model was better than the traditional model(Fuzzy AHP evaluation method)in Fengtai County.