Adaptive prediction model for fluidized catalytic cracking processes based on the PLS method

Abstract

Based on the partial least squares (PLS) method, an adaptive predictive PLS (AP-PLS) method was developed for sensitive adaptation to process changes using large-scale data (Big Data) from chemical processes. Utilizing data sets of fluidized catalytic cracking (FCC) and residue FCC (RFCC) processes as the basis, the AP-PLS method was developed and its prediction ability was compared with the simple PLS method. The required parameters for the prediction model of the FCC and RFCC processes are readily available from reported process data using time-varying model updating by the AP-PLS

Adaptive Modeling Strategy Integrating Feature Selection and Random Forest for Fluid Catalytic Cracking Processes

“…As the FCC reaction system is complex and the reactants and products are complicated mixtures, it is already difficult to determine all possible reactions, let alone the complex interdependencies between the reactions. Whereupon, data-driven approaches (also known as machine-learning methods) were developed, which required the harvesting and utilization of “big” process data and effective machine learning algorithms to model an FCC process and perform targeted prediction tasks, regardless of the complicated physicochemical mechanisms. − Ge presented a data processing and modeling procedure that includes data preprocessing, variable selection, model training, and other steps and reviewed the plant-wide process monitoring methods such as multiblock principal component analysis (PCA) and multiblock partial least squares (PLS). Later on, four main probabilistic latent variable models covering probabilistic PCA, factor analysis, probabilistic PLS, and probabilistic independent component analysis (ICA) for the high-dimensional nature of data in the process industry were summarized .…”

“…15 Jiang et al 16 presented a model based on a generalized regression neural network and an adaptive boosting algorithm for the prediction of gasoline yield. Kim and Lee 17 introduced an adaptive predictive-partial least squares (AP-PLS) method using big process data to cope with the process changes in FCC and residue FCC (RFCC) processes. However, the laws embodied in data-driven methods are often difficult to give a clear explanation, which is usually beyond the scope of the existing production technicians' understanding of FCC.…”