Extended period time series prediction of adaptive gray-box fuel consumption for variable pitch ships based on ET-Informer

Abstract

This paper describes an extended period time series gray-box fuel consumption prediction algorithm for variable pitch ships based on Event-Triggered Informer (ET-Informer). A white-box fuel consumption model is built by modeling ship resistance as a function of speed and pitch, then calculating the corresponding shaft power. An alternate approach is to use an innovative ET-Informer black-box algorithm which could preserve critical data features, minimize redundancy, enhance computational efficiency, extract key data to mitigate interference, and achieve extended time-series predictions. The proposed adaptive gray-box model builds on both white-box and black-box approaches, incorporating an improved Newton-Raphson-Based Optimizer (NRBO), human experience coefficients, and the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm to dynamically adjust weight coefficients based on model validation. This gray-box approach accelerates computation and addresses issues of singularity frequently encountered traditional algorithms. The validation is carried out based on operational datasets obtained from typical vessel operations across key maritime corridors. The findings of the study demonstrate that the proposed model is effective in performing fuel consumption optimization, thus improving fuel efficiency and its potential for practical applications.