Online Adaptive Modeling Framework for Deep Belief Network-Based Quality Prediction in Industrial Processes

A Feedforward Model Predictive Controller for Optimal Hydrocracker Operation

Hydrocracking is an energy-intensive process, and its control system aims at stable product specifications. When the main product is diesel, the quality measure is usually 95% of the true boiling point. Constant diesel quality is hard to achieve when the feed characteristics vary and feedback control has a long response time. This work suggests a feedforward model predictive control structure for an industrial hydrocracker. A state-space model, an autoregressive exogenous model, a support vector machine regression model, and a deep neural network model are tested in this structure. The resulting reactor temperature decisions and final diesel product quality values are compared against each other and against the actual measurements. The results show the importance of the feed character measurements. Significant improvements are shown in terms of product quality as well as energy savings through decreasing the heat duty of the preheating furnace.

Adaptive soft sensor based on transfer learning and ensemble learning for multiple process states

The objective of this study is to develop an adaptive software sensor technique that can predict objective process variables for a target grade in a plant while also considering information related to various other grades. We use a dataset of the target grade as the target domain and those of the other grades as source domains to perform transfer learning. Multiple models or sub-models are constructed by setting a source domain for each grade and changing the number of samples used as the source domain. Furthermore, to prevent the negative transfer, the use of a source domain is automatically judged. In this study, we constructed sub-models using the locally weighted partial least squares approach as an adaptive soft sensor technique. The values of an objective variable were predicted with ensemble learning using sub-models. The effectiveness of the proposed method was verified using a dataset measured in an actual incineration plant, and the proposed method was able to accurately predict the product quality even when the plant was operated in five grades and when a new grade was produced.