Machine learning-driven energy management of a hybrid nuclear-wind-solar-desalination plant

The ongoing energy transition and incoming water scarcity crisis demand coordinated research to ensure a fossilfree future for humankind. Aiming to increase energy efficiency, reduce curtailment and decarbonize water production, this paper proposes a novel energy management system (EMS) for a hybrid plant compound by a small modular nuclear reactor acting as cogeneration unit, a wind and solar farms as generators. Additionally reverse osmosis and multi-stage flash desalination plants are included as demand responsive units along with a freshwater storage. Mixed integer linear programming (MILP) is employed to formulate this stochastic optimization problem, where piecewise linear functions define operational costs and efficiencies of SMR and desalination motivating energy efficiency and safety. Renewable availability point forecasts are obtained with physics informed machine learning models whose error is characterised by fitting the predictor's residuals to different statistical distributions following an unsupervised methodology. The suitability of the EMS is addressed in two study cases, one exploring the flexibility exploitation of the algorithm and another proving its suitability for realtime implementation. The dispatcher manages to keep unaltered the SMR's core reaction while satisfying both electrical and water demand in different renewable availability regimes by fully exploiting sector coupling flexibility. Simultaneously, renewable curtailment is kept to a minimum.

Automated summary

This paper proposes a novel energy management system that utilizes a small modular nuclear reactor, wind and solar farms, reverse osmosis, and flash desalination plants to increase energy efficiency, reduce curtailment, and decarbonize water production. The system is optimized using mixed integer linear programming and renewable availability is predicted using physics informed machine learning models. The suitability of the system is validated through two study cases.