Local Search Techniques for Scheduling Problems : Algorithms and Software Tools

Local Search meta-heuristics are an emerging class of methods for tackling combinatorial search and optimization problems, which recently have been shown to be very effective for a large number of combinatorial problems. The Local Search techniques are based on the iterative exploration of a solution space: at each iteration, a Local Search algorithm steps from one solution to one of its “neighbors”, i.e., solutions that are (in some sense) close to the starting one. One major drawback of this family of techniques is the lack of robustness on a wide variety of problem instances. In fact, in many cases, these methods assure finding good results in reasonable running times, whereas in other cases Local Search techniques are trapped in the so-called local minima. Several approaches to the solution of this problem recently appeared in the literature. These approaches range from the employment of statistical properties (e.g., random explorations of the solution space), to the application of learning methods or hybrid techniques. In this thesis we propose an alternative approach to cope with local minima, which is based on the principled combination of several neighborhood structures. We introduce a set of neighborhood operators that, given a collection of basic neighborhoods, automatically create a new compound neighborhood and prescribe the strategies for its exploration. We call this approach Multi-Neighborhood Search. Although a broad range of problems can be tackled by means of Local Search, in this work we restrict our attention to scheduling problems, i.e., the problems of associating one or several resources to activities over a certain time period. This is the applicative domain of our research. We present the application of Multi-Neighborhood Search to a selected set of scheduling problems. Namely, the problems tackled in this thesis belong to the classes of educational timetabling, workforce assignment and production scheduling problems. In general, we obtain improvements w.r.t. classical solution techniques used in this field. An additional research line pursued in the thesis deals with the issues raised by the implementation of Local Search algorithms. The main problem is related to the difficulty of engineering the code developed for Local Search applications. In order to overcome this problem we designed and developed an Object-Oriented framework as a general tool for the implementation of Local Search algorithms. The framework, called EasyLocal++, favors the development of well-engineered Local Search applications, by helping the user to derive a neat conceptual scheme of the application and by supporting the design and the rapid implementation of new compound techniques. Therefore, EasyLocal++ allows the user to focus on the most difficult parts of the development process, i.e., the design and the experimental analysis of the heuristics.