Enhanced forecasting of shipboard electrical power demand using multivariate input and variational mode decomposition with mode selection

Accurate forecasting of shipboard electricity demand is essential for optimizing Energy Management Systems (EMSs), which are crucial for efficient and profitable operation of shipboard power grids. To address this challenge, this paper introduces a novel hybrid forecasting approach that combines multivariate time series decomposition with Machine Learning (ML) techniques. Specifically, the method utilizes Long Short-Term Memory (LSTM) networks to generate forecasts from multivariate input time series that have been decomposed using a newly formulated Variational Mode Decomposition (VMD), termed Variational Mode Decomposition with Mode Selection (VMDMS). VMDMS enables a selective detection process, identifying modes across channels that synergistically enhance forecasting accuracy. The proposed hybrid forecasting method is validated using a dataset of electric power demand time series collected from a real-world large passenger ship. Experimental results confirm the effectiveness of the approach, extending the applicability of VMD to multivariate forecasting without imposing restrictive assumptions on the data. This work contributes to ongoing efforts in optimizing decomposition methods for predictive modeling in energy management, opening new avenues for improving shipboard power grid efficiency.