论文部分内容阅读
多模型可以改善模型估计精度,提高泛化性能。针对传统的聚类方法过于依赖空间数据先验知识及初始参数的缺点,提出1种适用于任何形状样本分布的单参数调节扩张搜索聚类算法。该方法以近邻算法为基础,通过设定各样本的ε,邻域,以扩张搜索的方法将所有相关的ε-邻域样本归为一类,从而聚类样本数据。将其用于聚类样本数据集,构建基于扩张搜索聚类的软测量多模型。在双酚A生产过程质量指标的软测量建模仿真中验证了算法的有效性,其均方根误差、最大相对误差和平均相对误差均较基于模糊C均值的多模型建模方法有所减小,分别从1.2943,3.88%和1.40%下降到了1.0276,2.72%和1.16%。
Multi-model can improve the accuracy of model estimation and improve generalization performance. Aiming at the disadvantage that the traditional clustering method relies too much on the prior knowledge of the spatial data and the initial parameters, a new one-parameter adjusting extended search clustering algorithm suitable for any shape sample distribution is proposed. Based on the nearest neighbor algorithm, this method classifies all related ε-neighborhood samples into one category by setting ε and neighborhood of each sample and expanding the search method to cluster the sample data. It is used to cluster the sample data set and construct the soft-sensing multi-model based on the extended search clustering. The validity of the proposed algorithm was verified in soft sensor modeling and simulation of BPA production process. The RMSE, the maximum relative error and the average relative error were both lower than those based on fuzzy C-means Small, down to 1.0276, 2.72% and 1.16% from 1.2943, 3.88% and 1.40% respectively.