Groupwise evacuation with genetic algorithms

Abstract

In a crisis situation on board a ship, it can be of the utmost importance to have the passengers safely evacuate to the lifeboats in an efficient manner. Existing methods such as marked escape routes, maps and so on are not optimal as pre-planned escape routes may become heavily congested by passengers. The closest lifeboat is not always feasible as lifeboat capacity can be exceeded. Considering that some evacuees are strongly affiliated and would like to evacuate together as a group, it all becomes a very difficult problem to solve. Sub-problems have been modelled, but no existing model combines all of these aspects into account. We proceed by modelling the area to be evacuated as a time-expanded graph, assuming that future development in hazard severity is known in the form of a survivability percentage for each node. Then we apply a multi-objective genetic algorithm with five different fitness functions that use heuristics to maximize overall survivability and reduce the total egress time if possible. A method has been developed to pick the best evacuation plan out of the pool of potential solutions returned by the genetic algorithm. The solution is compared with Dijkstra’s algorithm and randomly generated paths. Experiments are conducted using these algorithms for both predefined and randomly generated graphs using different parameters. In the tested random graph, the genetic algorithm gives on average 24% better survivability and 3 times better grouping Random algorithms. A fixed network with a known solution was solved 100%. This genetic algorithm can be used to generate better routing plans that utilizes multiple evacuation routes and lifeboats while taking into account groups, resulting in smoother evacuations which can save more lives.