Energy management strategy for hybrid power ships based on nonlinear model predictive control

•An energy management strategy based on nonlinear model predictive control is designed for parallel hybrid power ships.•The real random waves model is established, and its effect on propeller load torque is observed by using an unscented Kalman filter.•The machine learning technology creates a representative test cycle for testing and evaluating the designed strategy.•A trade-off coefficient is designed to balance fuel consumption and carbon emissions. How to achieve energy saving and emissions reduction is a critical issue of great concern to all countries. The development of marine hybrid power technology provides a new research direction for rational energy distribution and utilization. In this paper, an energy management strategy for real-time regulation of internal combustion engines and motor torque is designed to solve the optimal energy management problem of parallel hybrid power ships. A nonlinear model predictive control method is used as the core, aided by a machine learning technique and an unscented Kalman filter. Machine learning techniques are used to extract representative test cycles and apply them to the testing and evaluation of design strategies. The effect of random waves on propeller load torque is observed using an unscented Kalman filter, which is used as a disturbance variable to participate in establishing the prediction model. Furthermore, a trade-off coefficient is designed to further demonstrate the balance between fuel consumption and carbon emission optimization. To verify the performance of the designed energy management strategy, NMPC is compared with the conventional intelligent optimization algorithm and dynamic programming algorithm, and the fuel consumption and carbon emission trade-off parameterization of the strategy is investigated. The results show that the strategy can resist random wave disturbances and follow the reference data of the ship’s internal combustion engine speed extracted by machine learning. The energy distribution is also performed according to the variation of the trade-off coefficients to achieve the optimal trade-off energy management under the required constraints.