(136b) Optimal Demand Response Operation of an Industrial Air Separation Unit Using Data-Driven, Scheduling-Relevant Dynamic Models

Renewable sources now make up a significant portion of the electricity generation portfolio, and the intermittency of such sources, coupled with market deregulation, has resulted in electricity prices that fluctuate over short time intervals (one hour or less). These time-sensitive prices effect demand-response behaviors, implicitly motivating consumers to lower electricity consumption during times of peak power demand, when electricity prices are typically also at a maximum. Chemical processes, in particular, are a natural candidate to benefit from such demand-side modulation of electricity use, as production rates (and electricity demand) can be increased during off-peak hours, and excess product can be stored and used to supplement lowered production rates during times of peak electricity demand [1]. In addition to lowering production costs, such demand-response (DR) modulation also has a stabilizing effect on the electricity grid, as electricity consumption is shifted away from times of peak demand to times of peak renewable-based generation, when electricity prices are low.

Air separation units (ASUs) use large amounts of electricity to separate air into purified gases via cryogenic distillation, with refrigeration cycles driven by large electric compressors. ASUs are a prime candidate for engagement in DR programs. Nevertheless, modulating air separation production rates in response to electricity prices involves operating in a transient regime, a clear deviation from the steady-state operational paradigms for which these processes are usually designed [1]. Additionally, dynamic models of air separation plants are usually nonlinear, high-dimensional, and ill-conditioned, making computation of optimal production schedules using dynamic models intractable in a practical amount of time [2]. In previous work [3], we took a data-driven approach, using historical operating data to build scale-bridging models (SBMs), or low-order models of the scheduling-relevant dynamics of a process. We showed that these models reduced the computation time required for the aforementioned scheduling calculations and that air separation units can improve their profitability by overproducing when electricity prices are high and storing oxygen, nitrogen, and argon in the form of cryogenic liquids.

In this work, we apply our framework for scheduling under dynamic constraints to an industrial ASU operating under Model Predictive Control, using historical industrial data to model closed-loop dynamics of the process. The process is a three-column ASU producing nitrogen, oxygen and argon, and we identify scale-bridging models of Hammerstein-Wiener form, that relate power demand to production rate targets and measurable disturbances (e.g., ambient conditions). We then solve scheduling problems using a sequential solver available in gPROMS, showing significant economic benefits compared to operation at a constant operating point. We also examine the increased benefits from scheduling a small network of multiple plants connected via a pipeline, which are subject to different time-varying electricity price trajectories, and we carry out sensitivity analyses to determine the effects of inaccuracies in electricity price forecasts on the economic benefits of the optimal production schedules.

[1] M Baldea and I Harjunkoski. Integrated production scheduling and process control: A systematic review. Comp. Chem. Eng., 71:377-390, 2014.

[2] LS Dias, RC Pattison, C Tsay, M Baldea, and MG Ierapetritou. Multi-resolution model of an industrial hydrogen plant for plantwide operational optimization with non-uniform steam-methane reformer temperature field. Comp. Chem. Eng., 107:271-283, 2017.

[3] RC Pattison, CR Touretzky, T Johansson, I Harjunkoski, M Baldea. Optimal process operations in fast-changing electricity markets: framework for scheduling with low-order dynamic models and an air separation application. Ind. Eng. Chem. Res., 55(16):4562-4584, 2016.