resolution methods have been proposed to deal with larger and more complex problems. Among those methods, metaheuristics are commonly used in academia and the industry to efficiently solve optimisation and constraint satisfaction problems. Nevertheless the design of increasingly efficient metaheuristics often leads to highly complex systems which require a non negligible amount of expert knowledge of the problem itself and of the parameterisation of the solving method. Designing automous search algorithms is thus an important topic. This thesis deals with the problem of managing and selection operators in the context of local search, within a generic controller. The latter should be easily adapted to deal with different problems. We therefore focus on designing simple and robust methods. Operator selection is based on learning the past performance of each operator to determine which operators are likely to be the most beneficial at each step of the search. In order to carry out this selection, the controller uses information about the capacity of the operators to improve the quality of solutions as well as their propensity to produce solutions which differ from ones already obtained. The proposed methods are tested on different theoretical problems and ones with practical applications. Both combinatorial optimisation and constraint satisfaction problems are considered. The results show that it is possible to obtain good results with simple methods. The proposed adaptive mecanisms are shown to be robust across different problems.
