多元回归及变量筛选表(LLJ)的内涵及应用

Connotation and Application of the Mulutiple Regression and Variables Screening Table of LLJ

借助微软的Excel平台,推出一种多元回归及变量筛选的实用科学方法-LLJ表.通过对超声回弹综合法检测混凝土强度模型的建立,展现LLJ表的灵敏性和实用性.以多元回归理论为基础,计算机软件Excel为平台,建立变量筛选LLJ表.经过LLJ表筛选后的变量所建的模型与原数据进行比较,拟合度良好.信度在变量筛选过程中的转加法则,其核心是在后退法任何一轮的筛选中没有达标(一般达标取:Bd≥0.90)的最小信度所对应的变量可永久性删除,并且只删除这唯一的变量;LLJ表是在插入函数LINEST基础上的扩展(由原来的5行扩展到9行);LLJ表对建立多元模型具有优越性、筛选过程中有极高的洞察力和对试验实现优化设计的特点.

By utilizing the Excel of Microsoft system, a practical method for multiple - regression and selecting variables is proposed, named Table of LLJ. According to the ultrasonic - rebound method for testing concrete strength of model, the Table of LLJ showed sensitivity and practicability. Comparing the model of variables selectded by the Table of LLJ with the original data, the degree of titting by Table of LLJ was excellent. The core of transferable adding regulation for reliability in the process of selecting the variables shows that the variables corresponding to the minimum reliability, which does not reach the standard and is selected at any round by backing method, can be cancelled permanently and exclusively. The common standard value of reliability is ≥0.09. The Table of LLJ is expanded by inserting the function of LINEST, which is expanded from 5 lines to 9 lines. The Table of LLJ features for the superiority for building multivariate model, and the selecting process features for better discenment and optimization design for testing parameters.