Ship energy management system development and experimental evaluation utilizing marine loading cycles based on machine learning techniques

In order to develop energy management systems for hybrid ship propulsion plants that are truly optimal and robust, it is important that the test conditions in experimental facilities are as close as possible to real world applications. In this context, a framework for the design and experimental evaluation of power-split control systems for ship propulsion is proposed. Using machine learning, data from ship operation are processed and 20 loading patterns are recognized; representative templates are extracted to be used as marine loading cycles in the energy management system development and testing. A ship propulsion model with wave disturbance is utilized to simulate realistic loading scenarios on the experimental facility. A predictive energy management system is presented, that controls the diesel engine and the electric motor/generator based on a strategy that defines the trade-off between fuel consumption and NOx emissions minimization. In addition the propeller load characteristics that are estimated and a speed predictor are utilized to aid the optimization within the 10 s prediction time window. A parametric simulation study is performed for the trade-off evaluation between fuel consumption and NOx emissions reduction potential of the control scheme. Finally, utilizing an extracted loading cycle, the energy management system is experimentally implemented and tested in real-time operation, where it has to cope with environmental disturbance rejection and follow the desired speed profile while performing the power-split control in respect to the fuel to NOx weighting strategy. Based on the experimental results in a hybrid diesel-electric marine powertrain with a 260 kW diesel engine and a 90 kW electric machine, fuel consumption and NOx emissions reduction by 6% and 8.5% respectively, were achieved over the tested profile. In this framework, the capabilities of the energy management system in realistic operation conditions can be exploited and evaluated.

Automated summary

For the development of optimal and robust energy management systems for hybrid ship propulsion plants, it is important to simulate real-world conditions as closely as possible in experimental facilities. This study proposes a framework for the design and experimental evaluation of power-split control systems for ship propulsion. Machine learning is used to process ship operation data and identify 20 loading patterns, which are extracted as representative templates for the development and testing of the energy management system. A ship propulsion model with wave disturbance is utilized to simulate realistic loading scenarios. A predictive energy management system is presented, controlling a diesel engine and an electric motor/generator based on a strategy that minimizes fuel consumption and NOx emissions. Parametric simulations are performed to evaluate the trade-off between fuel consumption and NOx emissions reduction potential. Experimental results on a hybrid diesel-electric marine powertrain show reductions of 6% in fuel consumption and 8.5% in NOx emissions compared to the tested profile.