Working fluid selection for Organic Rankine Cycles based on continuous-molecular targets

Abstract

Organic Rankine Cycles (ORCs) use low-temperature heat to generate electrical power. To ensure optimal use of a heat source, the cycle needs to be tailored to the specific application. Tailoring the cycle means optimizing both process and working fluid. This leads to a mixed integer nonlinear program (MINLP) of prohibitive size and complexity. Today, the selection of working fluid and process optimization are typically carried out separately following a two-step approach: In a first step, working fluid candidates are preselected using heuristic knowledge; in the second step, the process is optimized for each preselected working fluid. If the heuristics underlying the preselection fail, the optimal working fluid is excluded and this approach leads to suboptimal solutions. Continuous-molecular targeting (CoMT) is a framework for simultaneous optimization of process and working fluid [1]. Herein, working fluid properties are calculated by a physicallybased thermodynamic model, the perturbed-chain statistical associating fluid theory (PCSAFT) [2]. A set of pure component parameters describes each working fluid. These pure component parameters are relaxed during the simultaneous optimization of process and working fluid. Relaxation transforms the MINLP into a nonlinear program (NLP). The solution is a hypothetical optimal working fluid and the corresponding optimal process. In general, the hypothetical optimal working fluid does not coincide with a real fluid. Thus, real working fluids with similar properties are identified in the following step, the so-called structure mapping. Currently, a Taylor approximation of the objective function around the hypothetical optimal working fluid is used to estimate the objective function value of real working fluids. The Taylor approximation does not account for changes in the active set of constraints, whereby a substantial deviation between the Taylor prediction and the actual performance can occur leading to poor classification of the real working fluids. We present an iterative method to improve the approximation in the structure-mapping. A Taylor approximation is added around a new sampling point if its prediction is poor. The Taylor approximations from different points are combined using inverse distance weighting. The starting point is the optimal hypothetical fluid identified in the simultaneous optimization. The result of the method is a ranked set of working fluids. The iterative method improves the quality of the ranking and allows for efficient identification of the best working fluids. The approach is demonstrated in a case study for working fluid selection of a solar ORC.