一种面向订单剩余完工时间预测的SOM-FWFCM特征选择算法#br#

准确的订单剩余完工时间预测有助于动态调整生产计划、优化制造过程，以满足订单产品按时交付的需求。订单剩余完工时间受到车间物料、设备、在制品等各类生产要素的综合影响，相关数据具有典型的大量、多维、高冗余的特点，有效的特征选择能够获得更高的预测精度。在构建候选特征集的基础上，提出了一种基于自组织映射（SOM）网络特征加权模糊C均值（FWFCM）的特征选择算法。通过构建SOM网络初始化FWFCM的聚类中心，减少后者对初始聚类中心的依赖；基于互信息计算特征权重，实现导向性特征聚类，根据聚类结果选择特征代表，构成高质量关键特征子集。以某机加工车间的生产数据为例，通过与其他4种特征选择算法的对比分析，验证了所提算法的有效性。

A SOM-FWFCM Based Feature Selection Algorithm for Order Remaining Completion Time Prediction#br#

Accurate ORCT prediction was helpful to adjust production schedule and optimize manufacturing processes dynamically, which ensured timely orders delivery. ORCT affected by various production factors, including materials, equipment, works-in-process, et al. The related data possessed typical characteristics of large-scale, multi-dimensions and high-redundancy. Effective feature selection might improve the prediction accuracy. On the basis of constructing candidate feature sets, a feature selection algorithm was proposed based on SOM-FWFCM algorithm. Firstly, the cluster centers of FWFCM algorithm were initialized by SOM network to reduce the reliance on initial cluster centers. Feature weights were calculated by mutual information to achieve feature clustering with guidance. Then, according to the cluster results, representational features were selected to build high-quality key feature subset. Finally, taking the production data of a machining shop as an example, the effectiveness of the proposed algorithm was verified by comparing with other four feature selection algorithms.