基于改进NSGA3的焊接缺陷评估特征选择

在基于多源信息融合的焊接缺陷评估中, 特征选择对提高评估精度与速度发挥着重要作用。多源特征集由电弧电特征与电弧声音特征组成, 其特点在于特征集中存在冗余与互补特征。因此, 本文提出一种基于改进非支配排序遗传算法-Ⅲ(non-dominated sorting genetic algorithm-Ⅲ, NSGA3)的多目标特征选择方法, 旨在从多源特征集中找到最优特征子集。该方法首先对特征集进行相关、冗余和互补特性分析, 再以冗余性最小, 相关性与互补性最大为目标建立多目标特征选择优化模型。并基于相关、冗余和互补评价函数提出一种新的变异算子来引导变异过程, 以减少无效特征的影响, 提高收敛效率。实验采用支持向量机作为学习器来验证学习效果, 结果表明, 所提方法与其他3种方法相比, 可以在特征子集维度和预测精度方面获得更好的性能。

Feature selection for welding defect assessment based on improved NSGA3

Feature selection plays an important role in welding defect assessment based on multi-source information fusion because of improved detection accuracy and speed. The multi-source feature set consists of arc voltage features and arc sound features and is characterized by the presence of redundant and complementary features in the feature set. Therefore, this paper proposes a multi-objective feature selection method based on the improved non-dominated sorting genetic algorithm-Ⅲ (NSGA3), aiming to find the optimal feature subset in a multi-source feature set. Firstly, the method analyses the feature set for correlation, redundancy, and complementary characteristics, and then builds a multi-objective feature selection optimization model with the objective of minimizing redundancy and maximizing correlation and complementarity. Then a new mutation operator is proposed to guide the mutation process based on redundancy and complementary evaluation functions to reduce the influence of invalid features and improve convergence efficiency. Support vector machine is used as learners to verify the learning effect in the experiments. The results show that the proposed method can achieve better performance in terms of feature subset dimension and prediction accuracy compared with the other three methods.