Real-time energy management for the multi-source hybrid propulsion ship based on optimized DP algorithm and ensemble neural network

In response to the need for efficient operation under complex and fluctuating navigation conditions, this paper presents a multi-source hybrid propulsion system designed to dynamically switch propulsion modes in accordance with varying sailing loads. A rule-embedded dynamic programming (REDP) algorithm is proposed to solve this problem. The non-dominated sorting genetic algorithm II(NSGA-II) is used to co-optimize the battery capacity and the key parameters of the REDP algorithm. The pareto solution set obtained comprehensively considers peak factors, energy storage costs, and operating costs to select the optimal solution. To facilitate the real-time application of the REDP algorithm, the Markov chain is utilized to generate a comprehensive set of navigation conditions. The energy management strategy(EMS) dataset is obtained using the REDP algorithm and operation conditions. An ensemble neural network(ENet) model is proposed to learn the inherent nonlinear relationships within the dataset, replacing the reverse solving process in the REDP algorithm, to achieve real-time application of the REDP algorithm. The final results demonstrate that the calculation results of ENet can effectively track the calculation results of REDP algorithm, with peak factor and operating cost errors of 5.1% and 0.17% respectively.