Forecasting in Industrial Process Control: A Hidden Markov Model Approach

Abstract

The forecasting of information in industrial process control can assist in predicting the occurrence of future events. This is useful when such events are undesired and may lead to costly shutdowns of an industrial process. Thus, the forecasting of information can assist in determining preparatory measures to be taken to mitigate or prevent the occurrence of undesired events. This paper addresses the forecasting of information in industrial process control using a hidden Markov model approach. It defines three types of information signals from data of industrial process control (namely, low-dynamic, fast-dynamic, and multi-level switching). The paper discusses the complete methodology for forecasting of the signals; namely, pre-processing of the data, detection of the explanatory variables, determining the order of the forecasting model, implementation of the forecasting model, and generation and validation of the forecasted information. In addition, the paper implements the proposed methodology for two different practical systems with adjusting the parameters of the prediction model to demonstrate its applicability.