Solving CSPs with evolutionary algorithms using self-adaptive constraint weights.

This paper examines evolutionary algorithms (EAs) extended by various penalty-based approaches to solve constraint satisfaction problems (CSPs). In some approaches, the penalties are set in advance and they do not change during a run. In other approaches, dynamic or adaptive penalties that change during a run according to some mechanism (a heuristic rule or a feedback), are used. In this work we experimented with self-adaptive approach, where the penalties change during the execution of the algorithm, however, no feedback mechanism is used. The penalties are incorporated in the individuals and evolve together with the solutions