No abstract

No abstract

This work deals with memetic-computing agentmodels based on the cooperative integration of search agents endowed with (possibly different) optimization strategies, in particular memetic algorithms. As a proof-of-concept of the model, we deploy it on the tool switching problem (ToSP), a hard combinatorial optimization problem that arises in the area of flexible manufacturing. The ToSP has been tackled by different algorithmic methods ranging from exact to heuristic methods (including local search meta-heuristics, population-based techniques and hybrids thereof, i.e., memetic algorithms). Here we consider an ample number of instances of this cooperative memetic model, whose agents are adapted to cope with this problem. A detailed experimental analysis shows that the meta-models promoting the cooperation among memetic algorithms provide the best overall results compared with their constituent parts (i.e., memetic algorithms and local search approaches). In addition, a parameter sensitivity analysis of the meta-models is developed in order to understand the interplay among the elements of the proposed topologies.

The tool switching problem (ToSP) is well known in the domain of flexible manufacturing systems. Given a reconfigurable machine, the ToSP amounts to scheduling a collection of jobs on this machine (each of them requiring a different set of tools to be completed), as well as the tools to be loaded/unloaded at each step to process these jobs, such that the total number of tool switches is minimized. Different exact and heuristic methods have been defined to deal with this problem. In this work, we focus on memetic approaches to this problem. To this end, we have considered a number of variants of three different local search techniques (hill climbing, tabu search, and simulated annealing), and embedded them in a permutational evolutionary algorithm. It is shown that the memetic algorithm endowed with steepest ascent hill climbing search yields the best results, performing synergistically better than its stand-alone constituents, and providing better results than the rest of the algorithms (including those returned by an effective ad hoc beam search heuristic defined in the literature for this problem).

This paper presents a parameterized schema for building memetic algorithms based on cross-entropy (CE) methods. This novel schema is general in nature, and features multiple probability mass functions and Lamarckian learning. The applicability of the approach is assessed by considering the Tool Switching Problem, a complex combinatorial problem in the field of Flexible Manufacturing Systems. An exhaustive evaluation (including techniques ranging from local search and evolutionary algorithms to constructive methods) provides evidence of the effectiveness of CE-based memetic algorithms.