Dynamic Scheduling: A Comparison of High-Fidelity Models with Local Optimization versus Surrogate Models with Global Optimization

Flexible process
operation supports demand-side management under
fluctuating electricity prices, with integrated scheduling and control
enabling coordinated, real-time decisions. This work investigates
the trade-off between model fidelity and optimization complexity in
dynamic scheduling under integrated dynamic scheduling, using an air
separation process in a day-ahead electricity market. We explore mechanistic
and typical surrogate dynamic models yielding established optimization
formulations combined with standard global and local solution strategies.
(Non)linear surrogates are used in full-discretization, formulating
(mixed-integer) linear and nonlinear scheduling problems. Local dynamic
scheduling with full-order mechanistic models yields superior performance.
Conversely, deterministic global dynamic optimization incurs infeasibilities
for linear surrogates, and computational intractability for nonlinear
surrogates due to branch-and-bound limitations. Instead, stochastic
local optimization (multistart) with nonlinear models offers an alternative,
delivering high-quality solutions. We conclude that local dynamic
scheduling with mechanistic models, and multistart with Hammerstein-Wiener
models are preferable for available detailed process models or operational
data, respectively.