利用神经网络改进鸡疾病临床诊断专家系统  

Using Neural Network to Improve Expert System about Clinical Diagnosis of Chickens Disease

作　　者：陆昌华[1,2,3] 胡肄农[1,2,3] 陆庆文 王立方[1,2,3] 王启明[1,2,3] 

机构地区：[1]江苏省农业科学院农业现代化研究所 [2]南京大学 [3]江苏省农业科学院畜牧兽医研究所

出　　处：《江苏农业学报》1999年第1期42-46,共5页Jiangsu Journal of Agricultural Sciences

基　　金："九五"国家科技攻关项目;"九五"江苏省自然科学基金

摘　　要：探讨利用人工神经网络改进“鸡常见疾病临床诊断专家系统ＥＳＣＣＤ”。ＢＰ训练算法可解决专家系统构造中的瓶颈问题——知识获取，但存在收敛性和泛化能力较差的缺陷。作者使用比例训练的ＢＰ算法，提出对训练模式进行样本重组的方法，其特点是训练速度快、特征抽取能力强。结果表明，改进后的系统优于常规ＢＰ算法：系统可从病历中学习规则，诊断符合率从８０％提高到９４．１％，训练次数从７９００次（常规ＢＰ算法）减少到８００次，从而为探讨实现专家系统和神经网络的综合集成奠定基础，对构造类似专家系统具有普遍的适用意义。The improvement of Expert System about Chickens Common Diseases(ESCCD) using Artificial Neural Network was dicussed. BP training algorithm can resolve the bottle neck problem in building an expert system——knowledge acquisition, with the drawback of poor convergence and generalization. A scale training algorithm of BP neural network is used, and sample reorganization method is proposed. Its advantage is the fast training speed and good feature extraction ability. The results confirm that the modified ESCCD is better than conventional BP method:rules may be achieved by learning cases,congruence ratio is raised from 80% to 94 1%, and training epochs is reduced from 7 900 to 800. Thus,the foundation to carry out the integration of expert system and neural network is established, it has universal usefulness in building similar expert system.

关 键 词：神经网络 鸡病 诊断 专家系统 比例训练 

分 类 号：S858.314.4[农业科学—临床兽医学]