User-derived mutation in highly constrained truck loading optimization

Truck Loading Optimization problems from practice tend to involve a lot of constraints. In automatically solving these problems, a rule set has to be compiled that governs the generation of valid solutions. This rule set is either defined manually, or distilled automatically by analyzing solutions of human planners. This paper describes the extension of an existing optimization approach with statistics from man-made solutions through a so-called informed mutation operator. Solutions are scored on quality and on the percentage of violation, i.e., a combination of stacked boxes that did not occur in a separate set of man-made solutions. Applying the informed mutation operator decreases the average violation percentage in solutions, as well as achieving improved solution quality through fitting more boxes into a container.