Spatial reasoning with genetic algorithms an application in planning of safe Liquid Petroleum Gas sites

Abstract

The layout of a Liquid Petroleum Gas (LPG) site is subject to a number of constraints in terms of safety. Most of these are restrictions on the minimum distance between components of the site, such as loading bays, cylinder stores, tanks, buildings, roads, and the perimeter of the site, thus preventing small incidents triggering a dangerous sequence of events. Laying out such sites involves careful checking of distances and other safety rules, and Shell Research have developed a drawing tool which incorporates such checks. An experiment with genetic algorithms was undertaken to attempt the automatic generation of good, safe designs. The first attempt, reported here, designs cylinder stores. A cylinder store consists of a number of cylinder stacks represented as a set of rectangles of given area, which have to be separated by a minimum gap. The objective is to place all the stacks in a given area, of arbitrary shape, without contravening the safety distances. The genetic algorithm was able to produce interesting and safe layouts, and merits further investigation for more complicated problems. The genetic algorithm used involved a local hill-climbing step to speed convergence on solutions. The fitness function was based on a concept of repulsion and attraction between objects. The approach has generic applicability to design which involves the placement of components in two- or three-dimensional spaces.