基于非线性PLS回归的军用软件成本估算

当变量间存在多重相关性时，常采用偏最小二乘（partial least squares, PLS）回归进行建模。但是，传统PLS回归采用线性关系式来建立内部成分与外部成分之间的关系，已有的非线性PLS在模型可解释性方面存在不足。针对军用软件成本估算问题的非线性特点，建立了一种基于内部机理的非线性PLS回归模型，给出了模型的推导过程和实现算法。实例分析表明，该方法的估算精度优于多元线性回归，线性PLS回归和基于多项式内部映射的非线性回归。

Military software cost estimation based on nonlinear PLS regression

For the multi correlation of variables, partial least squares (PLS) regression is often adopted for modeling. However, traditional PLS establishes a linear relationship between inner and outer latent components, and existing nonlinear PLS models have some weakness in interpretability. Aiming at the nonlinearity in military softw are cost estimation, a nonlinear PLS regression model based on inner mechanism is established, and the derivation process and the arithmetic are presented. The experimental results show that the estimation accuracy is superior to multiple linear regression, linear PLS regression and polynomial nonlinear PLS regression.