Performance Analysis of GA and PSO Algorithms in Training Phase of Artificial Neural Network Model for Estimating Main Engine Power of LPG/LNG Ships

The production of LPG/LNG ships has also increased with the increasing demand for alternative energy sources such as LPG/LNG in countries. These ships, known as liquefied gas carriers, have their own characteristics and designs. In addition, they are designed by taking into account high safety standards because they carry dangerous cargo. Determining the required power for the main engine is one of the important steps in the initial stages of the design process. In this study, a main engine prediction model for LPG/LNG ships was produced using artificial neural networks (ANN). In the ANN training process, in addition to the basic backpropagation algorithm (BP) and Levenberg–Marquardt (LM) algorithms, heuristic algorithms, which have become increasingly popular in recent years and have been successfully applied in various disciplines, were also used. In this context, ANN training was also carried out with the most popular algorithms, Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). The main purpose of this study is to investigate the performance of heuristic algorithms in prediction model training. The results showed the superiority of the PSO algorithm among the intuitive algorithms. When comparing PSO with gradient-based algorithms, the PSO algorithm was superior to the BP algorithm, but performed worse than the LM algorithm. A global solution was obtained with the ANN model trained with LM, but the statistical analysis of the results revealed that the standard deviation of the LM algorithm was high. In contrast, the PSO algorithm consistently produced reasonable results with a lower standard deviation value. The Friedman test results also showed that the PSO algorithm would compete with LM.