【作者】 黄魏；

【导师】 吴礼树； 贺立源；

【作者基本信息】 华中农业大学 ， 植物营养学， 2007， 博士

【摘要】 在作物生长领域，营养元素作为生长的关键因子之一，当必需元素缺乏或过量时，作物会表现出特殊的症状。这种在外观上看到的症状有助于人们对作物营养状况的了解，也是诊断作物营养状态的重要指标。目前，在我国农村基层，还相对缺乏有经验的植物营养诊断专家；农民对作物因营养元素异常导致的症状表现认识不够，因此不能科学地诊断和及时校正营养元素的丰缺。研究将网络技术、数据库、自动分词等现代计算机技术与传统农业技术结合，开发面向专业领域的植物营养诊断自动应答系统诊断专家系统。研究分析和讨论了知识获取与知识表示的方法，在分析了植物营养知识特点的内在规律的基础上，采用了专家知识和文献知识两种主要获取方法。在研究过程中，参考了大量的相关资料，收集了覆盖常见作物的元素丰缺条件下的典型与非典型症状描述，直接进入数据库的有286条。以此为基础，提出了作物营养元素症状描述的客观标准集：1)按元素分类建立了12种元素的客观标准集；2)在对客观标准集分析归纳的基础上，建立了植株的表观症状描述、部位的表观症状描述、叶片的表观症状描述和诊断检索表；3)以SQL server2000为数据库平台，介绍了数据库的物理设计和逻辑设计。在已建各类知识的基础上，提出了面向植物营养诊断的分词模型PDCWSM(Plant Diagnose Chinese Word Segmentation Model)，从分词精度和可靠性判断，确定了以双向扫描法为主、逐词遍历法为辅的分词方法，对于典型症状描述，分词精度可达95％；对于非典型症状描述，分词精度可达80-85％，基本上满足了系统需要。推理过程结合植物营养专业知识和其内在规律性，提出了系统的回溯策略和不可回溯策略，提出采用正向推理结合反向推理的混合推理方式。根据植物元素功能及对症状的表现规律，研究建立了18条产生式推理规则，构建了CERTAIN与UNCERTAIN表，实现了在理想条件下三级推理和非理想条件的CF置信区间判断加提问的产生式推理过程。根据诊断结果建立了复混肥优化配方的对偶模型规划，利用其双向分析功能对配方进行了深入分析。在实现经济配方基础上，利用缩减成本分析了原料未采用的原因是其有效成分与价格不匹配。对已进入配方的原料进行了灵敏度分析，探讨了价格波动对配方的干扰程度，并指出磷元素为目前配方的价格最敏感因子。复混肥中三种主要元素的影子价格为钾＞磷＞氮，说明钾元素对总经济成本影响最大。更多还原

【Abstract】 Nutrition elements are one of the key factors in crop growing period. Crop showsspecial symptoms while lacking of essential elements or excessive. Those appearances ofsymptoms are important indicators to the crop nutritional status, which can helpunderstanding nutrition elements, are normal or not. At present, Chinese farmers havefewer agricultural knowledge and lack of correct guidance by experienced expert aboutplant nutrition diagnosis. They can’t judge the right condition and fertilize timely andpertinently. So we develop the web-based auto answering expert system of plant nutritiondiagnose, which integrate modern computer technology, network technology, database,automatic segmentation, and traditional agricultural knowledge.We adopt expertise and literature knowledge as two main knowledge acquire methods,based on discuss of knowledge acquisition and knowledge representation method, alsobased on analysis of plant nutrition function law and inherent features. We choose 286typical symptoms as system knowledge databases, after collected large amount of cropsymptoms knowledge. It covers common kinds crop. We bring forward the objectivecriteria sets: 1) 12 objective standards of symptoms of elements defect, 2) Apparentsymptom of plant, plant apparatus and leaf, diagnose search index 3) We use SQLServer2000 as database platform to analyse the physical design and logic design.We set up the PDCWSM (Plant Diagnose Chinese Word Segmentation Model) whichis adopted in the filed of plant nutrition diagnose. Judged by the accuracy and reliabilityjudgment, we use bidirectional scanning as main segmentation method and searchingword-by-word as secondly method. The segmentation accuracy is 95% for typicalsymptoms and 80-85% for untypical symptoms, which basically match the needs of thesystem requirement.An optimization dual model was developed to minimize cost over CompoundFertilizer and was described, which based on system diagnose results. The model wasused to illustrate the bidirectional effects of both raw material and constraints; especiallywhen the price of raw material was changing hotly. The most economic formulations werecalculated by Lindo 6.0. The reason of raw material unadopted was that their effective components were not match to price by Reduced cost analysis. The influence of adoptiveraw materials was discussed through Sensitivity analysis, and the phosphorus was themost sensitive factor in actual formulations. The sequence of main three elements’shadow price was K＞P＞N, which indicated the Potassium influenced the total costmostly.更多还原