Today, many algorithms are developed, evaluated, and compared on test benchmark problems. Their drawback is that they can simulate real-world problems up to some degree. Often, it turns out that there are many specifics of the problem we are trying to solve. To make algorithms efficient, constraints need to be considered and included in the problem solving. In this paper a real-world production planning problem is addressed. A typical approach with genetic algorithm turned out to be insufficient due to complexity with many constraints. To successfully solve this problem, a memetic algorithm, which uses specialized local searches to improve solutions acquired by genetic algorithm, is proposed. It is shown, that the use of specialized local searches can significantly improve the convergence and efficiency of the algorithm.