Modelagem rigorosa em regime dinâmico e controle preditivo de sistemas de transporte de fluidos compressíveis integrados a compressores centrífugos

Abstract

Natural gas has emerged as an energy source among fossil fuels due to its greater energy efficiency and possibility to be distributed via a pipeline network. Even so, to make it more competitive and accessible, tools capable of reducing the energy consumption necessary for its transport are needed. For this, knowledge of pipeline compression systems, especially centrifugal compressors, is essential since this equipment has high energy consumption and influences the degrees of freedom of operation, limited by surge and stonewall lines. From such constraints, the surge line stands out because it is located at the points of highest isentropic efficiency of the compressor and limits the stability of the process. Thus, this work contributes to the theme by modelling a compressor map phenomenologically, encompassing the non-ideality of the gas. From this premise, the following contributions have been studied: (i) evaluate the impact of the phenomenological model of compression systems both in the transient simulation of the networks and in the control of the compression systems, (ii) analyze the influence of the state equations in the simulations, as well as their impact on the transient states of the compressor map, (iii), propose control strategies for compression systems both in lumped and distributed parameter systems with the non-linear prediction of the surge curve. The open-loop simulations showed a significant influence of the compression system modelling compared with the literature practice, mainly when evaluating the transient behaviour of the surge prevention line. Meanwhile, in closed-loop simulations, an infinite horizon model predictive control strategy based on the successive linearization of the non-linear model of the process was proposed. This strategy proved promising in controlling the non-linear system while keeping the process in safe conditions, both in the simulations in the lumped parameter model and the distributed parameter model, with an adequate computation time for in-loco implementations.