【作者】 汪明磊；

【导师】 余为一； 耿照玉；

【作者基本信息】 安徽农业大学 ， 预防兽医学， 2001， 硕士

【摘要】 本研究以神经网络和基于案例的推理等人工智能技术为研究方法，以兽医临床常见的21种鸡病的表现症状、病理变化为研究对象，系统地研究了鸡病种类与其症状、剖检变化之间的复杂的对应关系，初步建成了以神经网络阵列诊断为主，基于病例的推理诊断为辅的鸡病诊断专家系统。经测试，单独使用系统的“直接诊断”模式的完全正确率为53.2％，未诊出率为12.8％，其余34％可把疾病的范围缩小到包括实际疾病在内的2-3种；在采用基于病例的推理诊断时，实际疾病的匹配度排在前两位的比例为57.5％；如果同时使用系统提供的“直接诊断”、“分类诊断”、“鉴别诊断”及“CBR诊断”等多种诊断方法，可使诊断的完全正确率达到70.2％，未诊出率为12.8％，其余17％可把疾病的范围缩小到包括实际疾病在内的2-3种。这些均基本达到辅助兽医临床诊断的要求。以“鸡传染性法氏囊病”为实例，证明增加病例，可有效地提高基于病例的推理诊断的正确性。 该专家系统采用了多种计算机技术（诸如多媒体、超文本技术），使其图文并茂，界面友好，操作简便，系统知识库、病例库易于修改、扩充，同时采用超文本技术集成了一套养鸡与鸡病防治的知识浏览系统，供用户查询、学习，并制作了有关高产蛋鸡饲养新技术的VCD光盘。 研究表明，利用人工神经网络、基于案例的推理技术开发兽医专家系统，提高其实用性是完全可行的，关键在于提供优质的学习样本和适宜的学习算法及推理策略。更多还原

【Abstract】 In this paper the complex relationship between the twenty-one kinds of clinic common chicken’s diseases and their symptoms, lesions was studied with the neural network and the case-based reasoning (CBR) technology, and an expert system for diagnosis of chicken’s disease was finally built based on the neural network array principle and the standard cases on the basis of the CBR and their matching algorithm. When this system was tested in a simulated diagnosis with direct neural network diagnosis, there would be 53.2% and 12.8% at the rate of exactness and failure respectively, and 34% could be defined within 2 or 3 kinds of possible diseases. Testing of this system by CBR diagnosis, the actual cases, whose matching degrees lay in the first or the second place, could make up 57.5%. When the multiform diagnosis of "direct diagnosis", "sorting diagnosis", "identifying diagnosis" and "CBR diagnosis" were used in the system, the rate of exactness could be raised to 70.2%, failure reduced to 12.8% and other to 17% defined within 2 or 3 kinds of possible diseases. So the system could generally fulfill the requirement for the application in veterinary clinic diagnosis. It was found that the rate of exactness in diagnosis could be further raised if the disease cases were added into the system by taking IBD (infectious bursal disease) as an example.Because more computer technologies (multimedia, hypertext et al) were used, this system would be provided with a friendlier man-computer interface, a stronger capability and easier operation; its knowledgebase and case base could also be modified and enlarged. A browsing system of the knowledge of the chicken breeding and the chicken disease prevention was built with hypertext technology. In addition, a VCD of breeding laying chicken was also built.Result in this study showed that it was feasible completely to constructing veterinary expert system with neural network and CBR technology. The keys to building of a better system might be learning samples with high quality, fitting learning algorithm and reasoning method.更多还原