论文部分内容阅读
标准农田是耕地的精华,是确保国家粮食安全的关键。科学评价标准农田地力等级对标准农田培肥和土壤改良有着重要意义。将粗糙集(Rough Set,RS)理论和支持向量机(SupportVector Machine,SVM)相结合,提出了基于RS和SVM的标准农田地力等级评价方法,同时,利用遗传算法的并行搜索结构和模拟退火的概率突跳特性,提出了GASA优化SVM参数算法。该方法首先在确定标准农田地力等级评价指标的基础上,利用地力调查样本数据及传统的指数和法评价结果构建RS决策表,应用RS穷尽算法对决策表进行约简,剔除冗余的评价指标,然后用约简后的评价指标作为SVM的输入,运用GASA优化SVM参数算法对SVM进行训练,建立标准农田地力等级的RS-SVM评价模型。应用该方法对温州市鹿城区标准农田地力等级进行评价,与未用RS约简的SVM模型和BP神经网络模型评价结果进行对比,SVM模型和BP神经网络模型的输入指标数均为15个,其评价正确率分别为100%和90%;RS-SVM模型的输入指标数为14个,其评价正确率分别为100%,结果表明,该方法通过RS约简评价指标后,SVM评价精度并没有降低,但降低了SVM输入向量维数和计算复杂度,提高了训练效率;SVM用于标准农田地力等级评价,具有比BP神经网络更高的评价精度,可有效用于标准农田地力等级评价,为耕地地力评价提供了新方法。
Standard farmland is the essence of arable land and is the key to ensuring national food security. Scientific evaluation criteria Farmland level of standard farmland fertilization and soil improvement is of great significance. The combination of Rough Set (RS) theory and Support Vector Machine (SVM) is proposed to evaluate standard farmland fertility level based on RS and SVM. At the same time, the genetic algorithm based on parallel search structure and simulated annealing Probability of sudden jump characteristics, proposed GASA optimization SVM parameter algorithm. Firstly, on the basis of determining the evaluation index of standard farmland grading, this paper constructs the RS decision table using the data of the geostationary survey and the traditional index and evaluation results of the law, reduces the decision table by using the RS exhaustive algorithm, removes the redundant evaluation index , Then use the reduced evaluation index as the input of SVM, and use GASA to optimize the SVM parameter algorithm to train SVM, and establish RS-SVM evaluation model of standard farmland geotechnical grade. This method was applied to evaluate the standard farmland productivity of Lucheng District in Wenzhou City. Compared with the result of SVM model and BP neural network model without RS reduction, the input indexes of SVM model and BP neural network model were both 15, The accuracy of the proposed method was 100% and 90%, respectively. The input index of RS-SVM model was 14, and the correctness rate of RS-SVM was 100% respectively. The results showed that the accuracy of SVM evaluation But it reduces the dimension and computational complexity of SVM input vector and improves the training efficiency. SVM is used for standard farmland grade evaluation, which has higher evaluation accuracy than BP neural network and can be effectively used in standard farmland grade evaluation , Which provided a new method for the evaluation of arable land productivity.