中文摘要：

针对复杂工业过程中存在的数据非线性的问题,对基于数据局部特征的回归模型构建和软测量建模方法进行研究.基于邻 域保持嵌入（NPE）算法思想,利用数据间局部关系特征,建立多目标的回归优化函数,提出了基于局部的数据回归（ LDR）算法.该方 法基于数据的局部关系和邻域特征,在保留输入数据和输出数据局部特征的同时,获取数据间的最大相关关系.通过数据低维潜变 量获取数据的回归关系,并建立软测量预测模型.将模型应用于工业案例中,预估产品的质量和难以在线测量的关键变量.脱丁烷 塔的案例研究证明了所提出的方法在变量预测方面的有效性.与基于全局特征的软测量模型的对比分析结果表明,所提出的LDR在 获取非线性数据相关性和增强数据预测精度方面具有显著的改善效果.

英文摘要：

To solve the problem of data nonlinearity in complex industrial processes,the method of constructing regression model and soft sensing modelling based on local feature of data are researched. Based on the concept of neighbourhood preserving embedding （NPE） algorithm, the multi - object regression optimization function is established by using the local relational feature,and the local based data regression（LDR） algorithm is proposed. Based on the local relation and neighbourhood feature of data,the method makes the input data and output data keep the local features and obtains the maximum correlational relation of data. Through data low - dimensional latent variables, the regression relation of data nature is obtained, and the soft sensing prediction model is established. The model is applied in industrial case for predicting the quality of product and some of the critical variables that are difficult to measure on the production line. The research on the case of debutanizer column proves the effectiveness of the method proposed for variable prediction. Comparing with the soft sensing model based on global feature, the result shows that LDR can achieve significant improvement on prediction accuracy and getting data correlation for the nonlinear processes.