摘要
偏最小二乘回归(PLS)是有力的建模工具,其因子数的选取直接关系到模型的实际预测能力。在实际应用中,因子数的选取有几种常用的方法,并没有统一的标准。研究以烟叶为实验对象,采用留一交互验证法,建立总糖指标的定量模型。着重研究因子数对定量模型性能稳定性的影响,改进了现有的方法,并与目前常用的两种方法在模型性能及预测稳定性方面进行了对比分析。分析结果表明,改进的方法在降低模型复杂性的同时,模型的准确性也得到了相应的提高。
Partial least squares regression （PLS） is a powerful modeling tool. The selection of PLS＇s component number is directly related to the actual predictive ability of the model. Without a universally accepted standard, there are several common methods in practical application. A quantitative model about total sugar index is established, by setting tobacco as the research object, using leave-one-out cross validation method. The component number＇s influence on the stability of quantitative model is mainly studied in article. The existing method is improved, and comparing with two methods that currently used in the model performance and the stability of forecast. The result show that the improved method can not only reduce the complexity of the model but also increase the accuracy of the model.
出处
《计算机工程与设计》
CSCD
北大核心
2014年第5期1788-1791,共4页
Computer Engineering and Design
