The impact of approximate evaluation on the performance of search algorithms for warehouse scheduling

Abstract

The Coors warehouse scheduling problem involves finding a permutation of customer orders that minimizes the average time that customers' orders spend at the loading docks while at the same time minimizing the running average inventory, Search-based solutions require fast objective functions. Thus, a fast lowresolution simulation is used as an objective function. A slower high-resolution simulation is used to validate solutions. We compare the performance of a constructive scheduling algorithm to a genetic algorithm and local search approach. The constructive algorithm is based on a heuristic built specifically for this application. We also tested a hybrid of the genetic algorithm and local search approaches by initializing the search using the domain-specific heuristic. This hybrid genetic algorithm was able to find the best solutions when evaluated by the high-resolution simulation. Finally, we consider the effect of using the high-resolution simulation to filter a set of solutions found by the different approaches.Events do not occur at the same granularity in the two simulators. While the time units are the same in both simulators, the low-resolution simulation uses larger time steps between events than the high-resolution simulator. As a result, there will be differences in the running average inventory, I, as reported by the two simulators.-We had previously reported a 3 order of magnitude difference in the internal and external simulators [19]. The difference has been reduced to 2 orders of magnitude by using machines with considerably more memory 82 J. P. WATSON E¹ A¸. Figure 3. The external simulation values plotted against the internal simulation values for running average inventory Over the 30 runs in Figures 2 and 3, the genetic algorithm has generated 'best found' solutions that are very similar in evaluation on the internal simulator, but which display significant variation on the external simulator. This is typical behaviour for genetic algorithms when ran on noisy functions [30].When sampling a noisy function, it is very common for the signal-to-noise ratio to be such that the signal is increasingly lost in the noise as one samples better solutions that have increasingly 92 J. P. WATSON E¹ A¸.

A survey on metaheuristics for stochastic combinatorial optimization

“…The intuitive reason is that often in SCOPs there are many local optima, whose values may be also quite near, and in order to discriminate between local optima one needs that the estimation error is small with respect to the difference between the exact value of local optima. We have seen a practical confirmation of this in several experimental papers, for instance Watson et al (1999) and Costa and Silver (1998]. A more rigorous argument in favor of this requirement is that all metaheuristics with provable convergence properties need to use a number of samples increasing with the iteration counter.…”

“…A huge research area devoted to solving problems with simulated objective function is Simulation Optimization. Following the definition given by Fu (2003), Simulation Optimization means ''searching for the settings of controllable decision variables that yield the maximum or minimum expected performance of a stochastic system that is presented by a Bianchi et al (2004Bianchi et al ( , 2006Bianchi et al ( , 2002a, Guntsch (2003, 2004), Erel et al (2005), Liu (2007), Liu et al (2007) Gutjahr (2003, Birattari et al (2005), Watson et al (1999), Yoshitomi (2002), Yoshitomi and Yamaguchi (2003), Jellouli and Châtelet (2001), Gelfand and Mitter (1989), Gutjahr and Pflug (1996), Gutjahr et al (2000a), Roenko (1990), Fox and Heine (1995), Alrefaei and Andradóttir (1999), Alkhamis and Ahmed (2004), Alkhamis et al (1999), Homem-de-Mello 2000, Bulgak and Sanders (1988),…”

“…In fact, the few papers applying EC to SCOPs that we are going to outline below, either use very simple techniques for dealing with noise or rely on methods that are unrelated to the main EC literature on noisy fitness functions. Watson et al (1999) address a stochastic warehouse scheduling problem where the objective function must be estimated by simulation. The authors consider a GA, a solution construction heuristic specific for that problem, two local search and a random search algorithm.…”

Understanding Algorithm Performance on an Oversubscribed Scheduling Application

“…The original implementation of the crossover operator (with a variable number of selected position) was shown to work well not only for our domain but also for other scheduling applications (Syswerda, 1991;Watson, Rana, Whitley, & Howe, 1999;Syswerda & Palmucci, 1991). For our test problems, the results in this subsection show that the number of crossover positions influences the performance of Genitor, both in terms of best solutions found and in terms of the rate of finding improvements.…”

Static resource allocation for heterogeneous computing environments with tasks having dependencies, priorities, deadlines, and multiple versions

“…The development of heuristic techniques to find near-optimal mappings is an active area of research, e.g., [2,5,[7][8][9]13,14,31,32,15,23,33,36,38,41,52,54,56]. This study focuses on static mapping heuristics.…”