A Design Methodology Enabling the Efficient High-Fidelity Design of Combined Cycle Power Plants

Abstract

The demand for greater efficiency, lower emissions, and higher reliability in combined cycle power plants has driven industry to use higher-fidelity plant component models in conceptual design. Normally used later in preliminary component design, physics-based models can also be used in conceptual design as the building blocks of a plant-level modeling and simulation (M&S) environment. Although better designs can be discovered using such environments, the linking of multiple high-fidelity models can create intractably large design variable sets, long overall execution times, and model convergence limitations. As a result, an M&S environment comprising multiple linked high-fidelity models can be prohibitively large and/or slow to evaluate, discouraging design optimization and design space exploration. This paper describes a design space exploration methodology that addresses the aforementioned challenges. Specifically, the proposed methodology includes techniques for the reduction of total model run-time, reduction of design space dimensionality, effect visualization, and identification of Pareto-optimal power plant designs. An overview of the methodology’s main steps is given, leading to a description of the benefit and implementation of each step. Major steps in the process include design variable screening, efficient design space sampling, and surrogate modeling, all of which can be used as precursors to traditional optimization techniques. As an alternative to optimization, a Monte Carlo based method for design space exploration is explained conceptually. Selected steps from the methodology are applied to a fictional but representative example problem of combined cycle power plant design. The objective is to minimize cost of electricity (COE), subject to constraints on base load power and acquisition cost. This example problem is used to show relative run-time savings from using the methodology’s techniques compared to the alternative of performing optimization without them. The example additionally provides a context for explaining design space visualization techniques that are part of the methodology.