可再生能源驱动的膜技术：集成太阳辐照度预测用于光伏驱动苦咸水淡化系统的预测控制

Abstract Solar irradiance (SI) fluctuations disrupt photovoltaic (PV) power output, causing instabilities and unwanted shut-downs in directly-coupled PV-powered membrane desalination systems, reducing production rates, water quality, and energy efficiency. Conventional energy storage-based mitigation strategies increase costs and system complexity. Sky-imaging-based SI forecasting can analyse sky conditions and produce SI forecasts up to 15-min ahead, offering an alternative to minimise power fluctuation effects without extensive reliance on storage systems. In this study, an image-based SI forecasting system (SIFS) was integrated into a PV-powered brackish water desalination system. The SIFS employs a convolutional neural network-long short-term memory (CNN-LSTM) model, trained on images from a low-cost sky imager (KALiSI) to forecast SI values 2–15 min ahead. The forecasts were used to control a solenoid valve that temporarily bypasses the backpressure valve during periods of high sudden drops in PV power, often referred to as ramps, preventing pump shut-downs. System performance was experimentally evaluated under sunny, partly cloudy, and cloudy weather conditions. With the 5-min forecast, shut-downs on very cloudy days were reduced from 12 to two, increasing daily production by 5 %. On more challenging partly cloudy days, shut-downs fell from 11 to 9, with a 2 % production increase. Longer forecasting horizons further minimised shut-downs and optimised energy efficiency, with the lowest specific energy consumption at the 15-min horizon. Water quality remained consistent across all forecast horizons. The SIFS-based approach enhanced the PV-powered membrane system stability and efficiency, demonstrating the importance of predictive control strategies for mitigating shut-downs.