样本筛选与操作优化的可视化方法实现

Realizing Sieving Samples and Optimizing Operation by Visualized Method

可视化方法通过将多维空间数据映射到二维平面上 ,依据“物以聚类”属性对二维平面中各样本点进行分类、识别 ,找出离群样本并将其剔除掉 .与此同时 ,产生目标函数等值线 沿着等值线值增大或减小的趋势方向寻优 ,就很容易地确定出最优点或最优化操作区域 .以某卷烟厂生产烟丝为例进行分析和计算 结果表明 ,文中筛选样本数据方法中 ,离群点的剔除提高优化结果的准确度 ,最佳操作条件的确定可为工业生产或科学实验提供决策支持 .

For realizing sieving samples and optimizing operation by visualized method, the authors map the data of multidimensional distribution onto two dimensional plane; classify and recognize various sample points on two dimensional plane, find the outliers and reject them. Mean while, the authors get the contour line of objective function; and search along trend and direction of increase or decrease of contour line value by which the optimum point or optimized operating zone can easily be determined. The production of cut tobacco in a cigarette factory is taken as example. As shown by reckoning, the rejection of outliers in sieving data of samples improves the accuracy of optimized results, the determination of optimal operating condition provides industrial production and scientific experiment with support of decision making.