Real-time Energy Management of Low-carbon Ship Microgrid Based on Data-driven Stochastic Model Predictive Control

Abstract

With increasing restrictions on ship carbon emissions, it has become a trend for ships to use zero-carbon energy such as solar to replace traditional fossil energy. However, uncertainties of solar energy and load affect safe and stable operation of the ship microgrid. In order to deal with uncertainties and real-time requirements and promote application of ship zero-carbon energy, we propose a real-time energy management strategy based on data-driven stochastic model predictive control. First, we establish a ship photovoltaic and load scenario set considering time-sequential correlation of prediction error through three steps. Three steps include probability prediction, equal probability inverse transformation scenario set generation, and simultaneous backward method scenario set reduction. Second, combined with scenario prediction information and rolling optimization feedback correction, we propose a stochastic model predictive control energy management strategy. In each scenario, the proposed strategy has the lowest expected operational cost of control output. Then, we train the random forest machine learning regression algorithm to carry out multivariable regression on samples generated by running the stochastic model predictive control. Finally, a low-carbon ship microgrid with photovoltaic is simulated. Simulation results demonstrate the proposed strategy can achieve both real-time application of the strategy, as well as operational cost and carbon emission optimization performance close to stochastic model predictive control.