Particle swarm optimization-based collision avoidance

İNAN, Timur; BABA, Ahmet Fevzi · 2019 · Crossref

DOI: 10.3906/elk-1808-63

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical problem of vessel collision avoidance in ocean engineering, motivated by the fact that human error causes 84%–88% of tanker accidents and up to 96% of vessel collisions. To mitigate these risks, the authors propose an anticollision decision support system that integrates radar data, neural networks, fuzzy logic, and particle swarm optimization (PSO). The system aims to detect surrounding obstacles, predict their movements, assess collision risk, and generate safe, optimal avoidance routes. The methodology employs a four-component architecture. First, a radar simulation system, based on a mathematical vessel model, detects obstacles and records voyage information. Second, a nonlinear autoregressive (NARX) artificial neural network, trained using the Levenberg–Marquardt algorithm on recorded radar data, predicts the subsequent positions of vessels at 1-second intervals. The network achieved high accuracy, with regression values exceeding 0.997. Third, a Mamdani-type fuzzy logic system calculates the collision risk magnitude (0 to 1) using the distance to the closest point of approach (DCPA) and time to the closest point of approach (TCPA) as inputs. This assessment incorporates International Regulations for Preventing Collisions at Sea (COLREGs) to determine right-of-way. Finally, a PSO algorithm determines the shortest safe path by optimizing node priorities while avoiding a defined safe zone around target vessels. The system was implemented in MATLAB and tested across five scenarios involving single and multiple vessel crossings. Results demonstrated that the system successfully generated avoidance maneuvers compliant with COLREGs, such as altering course to starboard or maintaining steady course when appropriate. In comparative tests against genetic algorithms (GA) and ant colony optimization (ACO), PSO consistently produced shorter anticollision routes. For instance, with 10 individuals, PSO achieved a route distance of 2.3 miles compared to 3.56 miles for ACO and 2.4 miles for GA. Additionally, PSO required fewer iterations to converge on the best solution, averaging 10 iterations compared to 12 for ACO and 13 for GA. The significance of this work lies in demonstrating that PSO is more efficient than GA and ACO for generating optimal collision avoidance routes in terms of both path length and computational speed. The study concludes that the proposed system is feasible for real-world application if the neural network is trained with actual radar and GPS data rather than simulation data. Future work aims to incorporate environmental factors such as wind, waves, and currents to further enhance the system's realism and utility for maritime navigation.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction