Performance verification of the PSO-RF Grey Box Model for predicting ship Carbon Intensity Index (CII)

Abstract

To meet the international shipping emission reduction rules and rating requirements, the shipping industry has taken various measures to improve the compliance degree of greenhouse gas emissions. The Carbon Intensity Index (CII) is a core indicator in this process. Accurate prediction and performance analysis of the CII for ships play a crucial role in the shipping industry. However, the current research is facing the problems such as high computational costs and insufficient performance of the prediction models. This study utilizes the collected high-precision and high-resolution operational ship energy efficiency data, combined the law of energy transfer in ship machinery and the calculation regulations, and proposed a simplified calculation formula for the CII of operational ships, reducing the required parameters to one and significantly lowering the calculation difficulty and cost. Subsequently, this white-box model and the black-box model of random forest (RF) optimized by particle swarm optimization (PSO) were connected in parallel to obtain a CII prediction grey-box model. Performance analysis is conducted by assessing the applicability of different models for prediction, the quality of the input data, and the influence of various black-box models on the gray-box model. Compared with the WBM and BBM, the GBM is the best, R² has increased by 5.65% and 15.18% respectively, MAE has decreased by 0.0753 and 0.1138 respectively, RMSE has decreased by 0.074 and 0.1755 respectively, MAPE has decreased by 1.17% and 1.6% respectively, and the error range is more concentrated.