基于GSA的复杂产品关键质量特性识别

Critical-to-quality characteristics identification for complex products using GSA

为了识别复杂产品关键质量特性(critical-to-quality characteristics,CTQs),提出基于遗传模拟退火算法(genetic simulated annealing algorithm,GSA)的特征选择算法。所提算法将遗传算法(genetic algorithm,GA)与模拟退火算法(simulated annealing algorithm,SA)结合,兼有不错局部搜索与全局搜索能力。提出一种综合适应度函数应用于所提算法,以同时优化CTQ集分类性能和所选质量特性数。算例结果表明,所提算法能有效过滤无关、冗余质量特性,识别关键质量特性;与Memetic算法和信息增益(information gain,IG)算法相比,所提算法在识别更少关键质量特性的同时,得到更高预测精度。

To identify critical-to-quality characteristics （CTQs） for complex products, a genetic simulated annealing algorithm（GSA）based feature selection algorithm is proposed. As the proposed algorithm combines the genetic algorithm （GA） and simulated annealing algorithm （SA）, it has both good local search ability and good global search ability. Additionally, the proposed algorithm adopts an aggregated fitness function, which can optimize the classification performance on CTQ set and the number of selected quality characteristics simul- taneously. Experimental results illustrate that the proposed algorithm can efficiently eliminate irrelevant and re- dundant quality characteristics and identify CTQs, as it can identify fewer CTQs with even higher predictive accuracy compared with the Memetic algorithm and the information gain （IG） algorithm.