A study on the collision avoidance of a ship using neural networks and fuzzy logic

In this paper, the fuzzy inference system combined with an expert system is applied to collision avoidance system. Especially, calculation method of the collision risk by using neural network is proposed. At first, the membership functions of DCPA and TCPA are determined on the basis of simulation results using the KT equations. And then, the inference table is redesigned by using the ANFIS (Adaptive Network-based Fuzzy Inference System) algorithm. Secondly, additional factors, the ship domain, topological characteristics and restricted visibility, which can affect navigator's reasoning of the collision risk besides DCPA and TCPA are considered. Finally, MLP (Multilayer Perceptron) neural network to the collision avoidance system is applied to make up for fuzzy logic.

► At first, the designed fuzzy inference system considering the maneuverability of the own ship can replace the existing fuzzy system dependent on an interview. ► Secondly, the result of the neuro-fuzzy algorithm applied for improving the simply designed fuzzy system can smoothly infer the degree of the collision risk. ► Thirdly, the diverse parameters can be coincidently considered through the learning process using a neural network. ► Finally, this research appears to be a foundation for the ultimately economical and safe automation of machinery.