Hydrocracking is a complicated long-term process resulting from many coupled variables that affect manufacturing schedule and create loud noises. This paper presents a dynamic interval identification method for hydrocracking based on principal component derivative feature clustering to accurately identify the dynamic changes from data. Firstly, a principal component analysis (PCA) is used to extract the principal components of the key hydrocracking operating parameters. Then, the first-order derivatives are obtained from fitting polynomials of the principal components with sliding windows. After that, the K-means algorithm is used to identify the dynamic adjustment intervals for the principal component derivative feature clustering with the density peak technique used to determine the initial centers. The flexibility and effectiveness of this method are validated on an industrial petrochemical process. The results show that this method can avoid the influence of variable errors and accurately identify the dynamic adjustment intervals without priori knowledge.
摘要 加氢裂化过程流程长, 操作变量多且耦合严重, 加工方案多变, 数据存在大量噪声。为能准确地从数据中提取动态调整操作序列, 提出了一种基于主元导数特征聚类的加氢裂化动态调整区间识别方法。采用主元分析方法提取加氢裂化关键操作参数的主元, 再基于带滑动窗口的多项式拟合方法拟合主元数据, 提取拟合数据的一阶导数作为聚类特征, 设计基于密度峰值确定聚类初始点的K-means算法, 进行聚类分析, 从而识别出加氢裂化动态调整区间。中国某石化企业实际生产数据验证结果表明：该方法可避免单个或几个变量误差的影响, 能有效识别动态调整区间, 且不依赖先验知识。