A machine-learning approach to speed-up simulation towards the design of optimum operating profiles of power plants Academic Article in Scopus uri icon

abstract

  • © 2019 Association for Computing Machinery.Nowadays, liberalized energy markets give priority to power generation using renewable energy sources (RES) to minimize environmental impact and promote competitiveness. Demand changes and the variability caused by RES are two obstacles in achieving a stable electricity generation. In this context, operational strategies are the key to achieve a more stable and balanced energy generation. Such operational strategies are characterized by operation profiles which show valve operations that takes the plant from an initial state to a final state by means of a series of state variables transitions. Some work has been published in the literature that focuses on the coupling of simulation and optimization. However, this approach involves many iterations, demanding significant computation time. In this paper, a machine-learning approach is proposed that can be used to replace the rigorous simulation model with a surrogate model, which is obtained in short period of time and reduces dramatically the simulation time. The proposed approach has been initially tested in a case study that focuses on the generation of operation profiles of a hydraulic system composed of three interconnected tanks whose level changes are achieved by control valve manipulations and its operation is analogous to some power-plant components (e.g. drum boiler) in terms of valve manipulations.

publication date

  • March 16, 2019