(7d) Enhancing Ethylene Production Using Soft Sensors

The safe and profitable operation of any process operation relies on real time process information. The huge number of sensors in a chemical process often face the challenges of being in high temperature, pressure and corrosive environments, making online sensors subject to hardware failure and malfunction. In certain extreme situations measurement using hardware-based sensors is impossible. Soft-sensors can be used to cross-validate measurements and provide an alternative where it is impractical or impossible to utilize mechanical sensors. This paper and presentation present a “soft-sensoring” approach that has been effective on estimating outlet temperatures of cracking furnaces.

The idea of “soft-sensoring” is to define the correlation among measurements and map one group to another. Soft-sensor performance depends not only on the mapping algorithm, but also the measurements selected for modeling. Chemical process measurements are typically both abundant and redundant, which makes soft-sensoring both feasible and challenging. The choice of input measurements is crucial. Typically, variable selection methods such as normalized Mutual Information (MI) and Maximal Information Coefficient (MIC) are applied along with the Maximum Relevance and Minimum Redundancy (mRMR) algorithm. However, these methods still face limitations such as inadequate handling of common redundancy and inefficient ranking of variable importance. The result can be poorly performing soft sensors.

In this work, a mutual information-based relevance-redundancy algorithm is introduced for feature selection, in which the relevance and redundancy among process measurements are evaluated through a comprehensive correlation function and ranked according to their importance using a Greedy search algorithm to obtain the optimal variable set. This approach has been validated in both Ethylene and Delayed Coking furnaces to enable classical machine learning algorithms to accurately estimate the outlet temperature of the cracking furnace online. The results demonstrated significant reductions in both mean absolute error and root mean square error compared to other methods. The findings show that this approach can improve the reliability and efficiency of petrochemical production processes. Examples of where this has been effectively utilized and will be used going forward will be discussed.