Modular operation of renewable energy-driven reverse osmosis using neural networks for wind speed prediction and scheduling
Mohamed T. Mito, Xianghong Ma, Hanan Albuflasa, P.A. Davies
Abstract
Operating reverse osmosis (RO) systems with renewable energy (RE) can contribute greatly to water security. However, the stochastic and intermittent nature of renewables means that most large-scale RO relies on fossil fuels via a grid connection. Modular operation by connecting and disconnecting RO units is promising to power multi-unit RO entirely from RE. Nevertheless, it may lead to excessive start-ups/shutdowns, especially when using wind energy. This paper proposes using neural networks for wind speed prediction and scheduling to improve the modular operation of wind-powered RO. A modular operation technique was developed for a three-unit RO system with variable water output. To estimate the number of operating units, a neural network was designed to predict wind speed 24 hrs ahead, giving a correlation (R = 0.64) and a RMSE of 1.54 m/s against real data. Two approaches, high- and low-output scheduling, were defined to either maximise production or minimise unplanned shutdowns during modular operation. The high- and low-output scheduling reduced the number of start-up/shutdown cycles by 37.5 % and 75 % compared to unscheduled operation, leading to a 1.9 % and 2.3 % improvement in specific energy consumption, respectively. Overall, scheduled RO operation minimised unplanned shutdowns and delivered stable performance while following recommended operating procedures.