A Hybrid Ensemble Model for Interval Prediction of Solar Power Output in Ship Onboard Power Systems

Application of solar energy into ship power systems has been increasingly drawing attention. Accordingly, an accurate prediction of solar power plays a significant role in the shipboard power system operation. However, a photovoltaic (PV) generation system on the shipboard, different from the one on land, has to suffer more dramatic power fluctuations caused by weather variations and motions of the ships, which increase the uncertainty of PV power outputs. This paper proposes a hybrid ensemble method for optimal interval prediction of onboard solar power based on a stochastic ship motion model. A set of machine learning techniques are combined together with the particle swarm optimization (PSO) to constitute a hybrid forecasting model, including a back propagation neural network (BPNN), a radial basis function neural network (RBFNN), an extreme learning machine (ELM) and an Elman neural network. Furthermore, for different learning algorithms, an ensemble strategy is employed to reduce the forecasting error and various environmental variables along with ship moving and rolling impacts are taken into account. The developed model has been practically tested on a power system on a large oil tanker penetrated with PV energy and the data along the typical navigation route from Dalian in China to Aden in Yemen are selected for solar power prediction. The simulation results demonstrate its high accuracy, which provides a reliable reference for ship power system operators to achieve a better energy management.