Fuzzy C-means Cluster Based on Local Weighted Principal Component Regression for Soft Sensor of an Industrial Hydrocracking process

Soft sensors have been widely used to predict the quality of products in industrial hydrocracking processes. However, they are usually characterized with complex nonlinearities and multimodes. Hence, a single global model is often difficult to fully meet the requirement of prediction accuracy. In this paper, a novel fuzzy C-means cluster (FCM) based locally weighted principal component regression (LWPCR) is proposed for soft sensor modeling of an industrial hydrocracking process. First, FCM is utilized to cluster process data into different local clusters with similarity data patterns. When a query sample comes, LWPCR models are then built for accurate modeling in different modes with the most similar relevant samples. At last, the outputs of these sub models are fused to obtain a final estimation value of the query sample, in which the weights are automatically determined by the distances between the query sample and the cluster centers. The effectiveness of the presented approach is verified by predicting the 50% recovery temperature in an industrial hydrocracking process.