(203f) Optimal Multi-Scale Demand-Side Management for Continuous Power-Intensive Processes

Grossmann, I. - Presenter, Carnegie Mellon University
Mitra, S., Center of Advanced Process Decision-making, Carnegie Mellon University
Pinto, J. M., Linde plc

With the advent of deregulation in electricity markets and an increasing share of intermittent power generation sources, time-sensitive electricity prices (as part of so-called demand-side management in the smart grid) offer potential economical incentives for large industrial customers. These incentives have to be analyzed from two perspectives. First, on an operational level, aligning the production planning with the electricity price signal might be advantageous, if the plant has enough flexibility to do so. Second, on a strategic level, investments in retrofits of existing plants, such as installing additional equipment, upgrading existing equipment, or increasing product storage capacity, facilitate cost savings on the operational level by increasing operational flexibility.

We propose an MILP formulation that integrates the operational and strategic decision-making for continuous power-intensive processes under time-sensitive electricity prices. We demonstrate the trade-off between capital and operating expenditures with an industrial case study for an air separation plant. Furthermore, we compare the insights obtained from a model that assumes deterministic demand with those obtained from a stochastic demand model. The value of the stochastic solution (VSS) is discussed, which can be significant in cases with an unclear setup, such as medium baseline product demand and growth rate, large variance or skewed demand distributions. While the resulting fullspace optimization models are very large-scale, they can mostly be solved within up to three days of computational time.

Therefore, we also describe a hybrid bi-level decomposition scheme that addresses the challenge of solving the large-scale two-stage stochastic programming problem with mixed-integer recourse, which results from the multi-scale capacity planning problem. The decomposition scheme combines bi-level decomposition with Benders decomposition, and relies on additional strengthening cuts from a Lagrangean-type relaxation and subset-type cuts from structure in the linking constraints between investment and operational variables. The application of the scheme with a parallel implementation to an industrial case study reduces the computational time by two orders of magnitude when compared with the time required for the solution of the full-space model.