Risks Identification and Fire Scenarios Determination of Ship Fires Based on Improved Text Mining and K-Means Algorithm

Abstract

Ship fires pose significant threats to maritime safety. This study employs advanced text mining techniques alongside the K-means algorithm to develop a risk structure for ship fires, aiming to identify key risks and fire scenarios. We collected detailed fire investigation reports from authoritative sources, creating a database of 160 incidents over the past 20 years to analyze accident patterns. To enhance traditional text mining, we extracted 260 risk descriptors using specialized dictionaries, calculating their correlations. The improved K-means algorithm, utilizing cosine distance, clustered over 1000 related word combinations, revealing 13 key risks and 42 fire scenarios. From these findings, a risk structure was established through critical importance calculations. Results indicate that damage to flammable liquid tanks or pipes and improper storage of flammable solids are critical risks, elevating fire probability by over 15%. Risks like insulation failure and electrical short circuits showed critical importance values between 0.06 and 0.07. Notably, fire scenarios involving flammable oil leaks and electrical failures are interconnected, with the combination of flammable liquid leaks and insulation failure representing the most hazardous scenario, increasing fire probability by about 30%. This study introduces a data-driven approach to identify potential risks and fire scenarios, contributing practically to risk prevention and management in maritime accidents.