Evolutionary one-machine scheduling in the context of electric vehicles charging

Genetic Algorithm to Evolve Ensembles of Rules for On-Line Scheduling on Single Machine with Variable Capacity

“…From the above observations, it seems clear that it is not easy to evolve a rule that may generalize well over a large set of unseen instances so that it can compete with off-line algorithms such as GA introduced in [9]. Therefore, we propose considering ensembles or rules in the following way: each instance is Table 1.…”

“…It is important to remark that the N rules correspond to different generations of the GP, therefore their qualities are quite different. The set of M =50 instances is a subset of the set of 1000 instances proposed in [9]. The remaining 950 instances in this set are considered here to evaluate the best evolved ensembles.…”

“…We use here the schedule builder proposed in [9], which produces left-shifted schedules, a subspace of feasible schedules that includes at least one optimal solution. This schedule builder is depicted in Algorithm 1; it maintains a set U S with the unscheduled jobs, as well as the consumed capacity X(t) due to the jobs scheduled so far.…”

“…The schedule builder may be instantiated by using any priority rule or heuristic, as we will see in the next section. Besides, it could be embedded as a decoder in a genetic algorithm, as done in [9].…”

“…The results produced by the evolved ensembles are compared with those from the best on-line and off-line methods in the literature to solve the (1, Cap(t)|| T i ) problem. As far as we known, these methods are the GP proposed in [3] and the genetic algorithm proposed in [9]. The results of the experimental study show that the solutions obtained from the evolved ensembles are better that those produced by the sets of best rules obtained in [3] and that these solutions are actually close to those obtained off-line in [9].…”