In this paper, two studies are carried out related to the performance simulation and analysis of a wind-powered seawater reverse osmosis (SWRO) desalination plant prototype installed on the island of Gran Canaria (Spain). Three machine learning techniques (artificial neural networks, support vector machines and random forests) were implemented to predict the performance (pressure, feed flow rate and permeate flow rate, and permeate conductivity) of the SWRO desalination plant. Subsequently, plant operation was analysed in two different operating modes: a) constant pressure and flow rate through connection with a wind-battery microgrid, b) variable pressure and flow rate as a function of the power supplied by a stand-alone wind microgrid without energy storage. The paper supports two main outcomes. First, support vector machines and random forests are significantly (5% significance level) better predictors of the plant's performances than neural networks. Second, over one year, the operating mode that considers variable pressure and flow rate operates more continuously (higher operating frequencies and lower stop/start frequencies) than the constant pressure and flow rate alternative; however 1.2 times less permeate with 1.08 higher conductivity is produced on an annual basis.