经验公式发现系统(FDD)中对误差的改进  

Improvement of error in formula discovery from data

作　　者：冯金花[1] 陈燕雷[1] 关永[1] 

机构地区：[1]首都师范大学信息工程学院,北京100037

出　　处：《计算机工程与设计》2008年第20期5287-5289,共3页Computer Engineering and Design

基　　金：北京市自然科学基金项目(4062009);教育部重点基金项目(DF2006);北京市教委重点基金项目(KZ200710028014);北京市教委科技发展计划面上基金项目(KM200610028013);北京市教委教学改革立项基金项目(2005075)

摘　　要：在研究科学和解决工程应用问题时,经常需要根据两个变量的实验数据,找出这两个变量之间的关系。使用传统的数据拟合技术所求得的近似公式大多表示为代数多项式,系数由最小二乘法原理建立正规方程组求出。但这种传统拟合方法存在一个"病态问题",即系数行列式元素微小变化引起解的显著变化的问题。为了解决这个问题,采用基于人工智能的机器发现和数值计算的曲线拟合相结合的经验公式发现技术,并对经验公式发现系统中的误差评判方法提出了改进算法,提高了公式发现系统的可用性。In scientific research and solving engineering problem, it often needs to find out the relation between two variables according to the experiment data. The result of the traditional data fitting method are almost algebra polynomial, their modulus are decided from least square method. But this method has an ill-conditioned problem, which is it can bring large change in polynomial＇s solution because of little change in polynomial＇s elements. In order to solve this problem, the technology of formula discovery from data that based on machine finding of artificial Intelligence and the curve fitting of numerical calculation to solve this problem is used. An improving algorithm for the judging means to judging error in formula discovery from data is put forward, the usability of this system is increased.

关 键 词：经验公式发现系统FDD 数据拟合 最小二乘法 误差 平均误差率 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]