An ensemble of Extreme Learning Machine for prediction of wind force and moment coefficients in marine vessels
Krishna Kumar N, R. Savitha, Abdullah Al Mamun · 2016
In the recent times, offshore activities are getting increasingly important, and marine vessels are prevalent in all the water bodies. This requires a detailed study of the effect of environmental forces of the marine structures. This paper aims at developing an unified framework to study the effect of wind force and moments on marine vessels. A neural network approach is developed to study the effect of longitudinal and side forces of wind, and the yaw moment. The study considers various types of marine vessels at different loading conditions, with a total of 22 marine vessels. Of these, 18 are used to train an ensemble of Extreme Learning Machine (ELM) neural network. The network thus developed is tested for generalization on 2 new type of vessels at 2 different loading conditions. Thus, the developed model is capable of predicting the wind force and moment coefficients, irrespective of the type of vessel used. An Ensemble of extreme learning machine, each with input parameters initialized at different regions of the input space, are trained with all training samples. For each sample, the ELM that produces the least mean square error is identified, and the output of that ELM is considered as the output for that sample. Thus, the randomness of the initialization in ELM is exploited to achieve superior generalization performance. Performance study to predict the wind force and moment coefficients of marine vessels show that the ensemble of ELM has superior prediction performance, compared to state of the art results for this problem.