Machine learning assisted water management strategy on a self-sustaining seawater desalination and vegetable cultivation platform

Abstract

To enable vegetable cultivation in conditions lacking freshwater and soil, such as islands and offshore platforms, this study proposes to combine interfacial solar evaporation with hydroponic techniques to set up an electricity-free self-sustaining desalination and cultivation platform. To maintain a dynamic balance between freshwater supply and demand on the platform, we investigate the relationship between water consumption in hydroponic lettuce cultivation and meteorological conditions such as temperature, humidity, and solar radiation. The limitations of traditional evapotranspiration models like the FAO-PM model in practical hydroponic cultivation is highlighted and four machine learning algorithms, including Random Forest, Backpropagation Neural Network, Support Vector Regression, and N-BEATS, are developed to predict water consumption specifically for hydroponic cultivation. The Random Forest model, selected based on performance comparison, achieves MAPE and R2 values of 10.05% and 0.974 respectively on the test set. Using meteorological data from Shanghai in 2022, the feasibility of the platform is assessed, demonstrating a theoretical lettuce yield of 260 plants/m2/year for solar stills on a unit area without water shortages throughout the year. This study verifies the feasibility and stability of the hydroponic cultivation platform, and is promising for the development of sustainable agriculture in offshore scenarios with limited freshwater and soil resources.