本文介绍了如何使用MATLAB的GA-PLS方法进行特征波长的筛选。遗传算法能提取出连续且较少分散的波长，减少过拟合风险。通过随机化测试和进化代数优化测试确定最佳运行条件，然后通过多次运行遗传筛选确定变量数量。最终，使用选定的变量进行PLS模型的建立和预测。
常规特征提取的方法所提取的波长一般是分散的. GAs, after suitable modification, produces more interpretable results, since the selected
wavelengths are less dispersed than with other methods. GA可以提取出尽可能连续的波长。
遗传算法提取波长通常假定变量之间存在自相关.然而实际的波长变量之间不是如此的.That means that if wavelength n is selected as relvant,
wavelength n-1 and n+1 should also have a high probability of being selected.
GA的过拟合风险：随着测试变量数的增加而增大，因为变量数增加时，模型的良好性能更趋向于源于变量内在的随机相关关系。
To reduce the risk of overfitting, the final model is obtained from the results of 100 independent, very short GA runs.
it's strongly suggested to have a num
最低0.47元/天 解锁文章
扫一扫
2198
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
