Intelligent ethylene production relies on the coordination
of advanced
optimization techniques at different decision levels. For ethylene
cracking, the occurrence of dynamic drift and the semi-continuous
operation mode mean that decision variables at the operation and scheduling
levels are coupled and jointly determine the system performance. Therefore,
a collaborative operation and scheduling optimization framework is
required to improve efficiency and profitability, but this is difficult
to develop because of the coupled process variables. This paper proposes
a novel decoupling strategy addressing the two-level ethylene optimization
problem of batch operation and cyclic scheduling. Specifically, the
operation-level optimization consists of a set of nonlinear programming
(NLP) models that run in parallel with different feeds, furnaces,
and batch run lengths. The resulted optimal batch operation subsequences
and average yields are summarized into parameter tables and passed
to the upper mixed-integer nonlinear programming (MINLP) model, which
further determines the optimal feed allocation sequences and batch
timings at the scheduling level. When applied to an industrial scheduling
case, the two-level model improves the batch operations significantly,
increasing the ethylene and propylene yields by 0.02–1.24 wt
% and lengthening the maximum cracking time by 6–9 days for
all feed types. Compared with the traditional scheduling model in
which the coil outlet temperature (COT) is fixed, an increase of 2.5%
in the profit margin is achieved with the use of a multi-time period
operation model which allows the COT to vary and the maximum cracking
days to lengthen, demonstrating the significant potential of the proposed
two-level model for increasing ethylene cracking profit margins.