基于层次分析模糊模型与BP神经网络的凤台县耕地地力评价研究

发布时间：2018-06-13 22:20

本文选题：凤台县 + 地力评价　；参考：《安徽农业大学》2015年硕士论文

【摘要】：从1984年开始,农业部就在全国开展地力监测和评价工作,耕地地力评价是耕地质量的综合反映,提高评价精准,使其应用价值更高,对农业生产据有重要意义。本文以淮河平原凤台县为例,收集其专题图数据和文本数据,将专题图矢量化并在软件中叠置分析,形成综合耕地单元。将文本数据中异常值剔除,在ArcGIS中转化为点,并插值分析将所有养分数据和土壤属性数据赋给耕地单元,建立属性库。对于含有属性库的耕地单元采用不同方法进行耕地地力评价,并比较评价结果精准,为此分别通过传统方法层次分析模糊评价法和人工智能BP神经网络评价法评价,比较评价精度。首先以通过传统方法层次分析模糊评价法对耕地单元进行评价,选取与凤台县耕地息息相关的评价指标,分别为成土母质、地形地貌、耕层质地、耕层容重、CEC、pH、有机质、有效磷、速效钾、有效锌、有效铁、有效锰、剖面构型、耕层厚度、灌溉条件、排涝条件16项评价指标。建立层次模型,通过层次总排序算出每个评价指标的权重。再对每个指标拟合出相应的隶属函数,赋予其隶属度,最后累加得出每个耕地单元的评价得分,根据全国耕地地力等级分级标准分为五个地力等级,得到评价结果,其中尚塘乡、朱马店镇、杨村乡、古店乡和钱庙乡一二等耕地占这些乡镇总耕地面积的90.2%。其次再次通过人工智能神经网络对凤台县耕地单元再次评价,输入数据为16个评价指标,输出数据为常年目标单产。选取岳张集镇、顾桥镇、桂集镇、城北镇、新集镇、刘集乡、城关镇、经济开发区和李冲乡为训练样本数据,大兴集镇、关店乡和丁集乡为仿真验证数据,尚塘乡、朱马店镇、杨村乡、古店乡和钱庙乡为预测数据。选择tansig为输入函数和purelin为传递函数,trainlm为训练函数建立神经网络模型。隐含层节点数为14,训练精度达0.93以上,已建网络能作为预测模型。用建立好的网络模型预测尚塘乡、朱马店镇、杨村乡、古店乡和钱庙乡五个乡镇的耕地单元耕地地力。依据全国耕地地力标准划分,其中尚塘乡、朱马店镇、杨村乡、古店乡和钱庙乡一二等耕地占这些乡镇总耕地面积的91.3%。最后,把两种评价方法的评价结果与真实常年地力结果进行比较。按乡镇统计不同等级地力的数量比例,并绘制出耕地地力空间分布图。最终结果表明神经网络评价法的一二等耕地的相对误差为0.04和0.06,均小于传统方法的0.20和0.11,在各个乡镇的空间分布精度上神经网络的评价结果较传统评价更接近真实分布结果。研究表明在凤台县耕地神经网络模型评价结果要比传统方法层次分析模糊评价法评价结果优越。
[Abstract]:Since 1984, the Ministry of Agriculture has carried out the monitoring and evaluation of land fertility in the whole country. The evaluation of cultivated land fertility is a comprehensive reflection of cultivated land quality, which improves the accuracy of evaluation and makes its application value higher, which is of great significance to agricultural production. This paper takes Fengtai County of Huaihe Plain as an example, collects its thematic map data and text data, vectorizes the thematic map and analyzes it in the software to form a comprehensive cultivated land unit. The outliers in text data are eliminated and transformed into points in ArcGIS. All nutrient data and soil attribute data are assigned to cultivated land unit and attribute database is established by interpolation analysis. For the cultivated land unit with attribute bank, different methods are used to evaluate cultivated land fertility, and the results are compared and accurate. Therefore, the traditional analytic hierarchy process fuzzy evaluation method and artificial intelligence BP neural network method are used to evaluate the cultivated land fertility. Compare the accuracy of evaluation. Firstly, the cultivated land unit is evaluated by the traditional method of AHP fuzzy evaluation, and the evaluation indexes which are closely related to the cultivated land in Fengtai County are selected as soil parent material, topography and geomorphology, topsoil texture, cultivated layer bulk density, pH value of CECU, organic matter, available phosphorus, etc. Available potassium, available zinc, available iron, available manganese, profile configuration, thickness of tilling layer, irrigation condition, drainage condition 16 evaluation indexes. The hierarchical model is established, and the weight of each evaluation index is calculated by the total ranking of the hierarchy. Then fit the corresponding membership function to each index, endow it with the membership degree, finally accumulate the evaluation score of each cultivated land unit, and divide it into five land fertility grades according to the grading standard of the national cultivated land fertility grade, and get the evaluation result. Among them, Shangtang Township, Zhumaadian Town, Yangcun Township, Gudian Township and Qianmiao Township occupy 90.2 of the total cultivated land area of these townships. Secondly, the cultivated land unit in Fengtai county is evaluated again by artificial intelligence neural network. The input data is 16 evaluation indexes and the output data is perennial target yield. Select Yuezhangji Town, Guqiao Town, Guji Town, Chengbei Town, Xinji Town, Liu Ji Township, Chengguan Town, Economic Development Zone and Li Chong Township as training sample data, Daxing Town, Guan Dian Township and Ding Ji Township as simulation verification data, Shangtang Township, Zhu Madian Town, Yangcun Township, Gudian Township and Qianmiao Township are forecasted data. The neural network model is established by selecting tansig as input function and purelin as transfer function trainlm as training function. The number of hidden layer nodes is 14 and the training precision is more than 0.93. The established network can be used as a prediction model. The cultivated land fertility of five townships, Shangtang, Zhumaadian, Yangcun, Gudian and Qianmiao Township, was predicted by using the established network model. According to the national cultivated land fertility standard, the cultivated land in Shangtang, Zhumaadian, Yangcun, Gudian and Qianmiao is 91.3 percent of the total cultivated land area in these towns. Finally, the evaluation results of the two methods are compared with the results of real perennial soil force. The spatial distribution map of cultivated land fertility was plotted according to the quantity ratio of different land fertility in villages and towns. The results show that the relative errors of the first and second grade cultivated land of the neural network evaluation method are 0.04 and 0.06, which are less than 0.20 and 0.11 of the traditional method. The evaluation results of the neural network are closer to the real distribution results in the spatial distribution accuracy of each township than the traditional evaluation. The research shows that the evaluation results of neural network model of cultivated land in Fengtai County are superior to those of traditional analytic hierarchy process fuzzy evaluation method.
【学位授予单位】：安徽农业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S158

【相似文献】