A Self-Adaptive Parameter Control Method Based on Entropy and Security Market Line for Scheduling Problems

Scheduling problems, as one of the classic combinatorial optimization problems are essential issues in many fields. Various meta-heuristic algorithms have been adopted to solve scheduling problems. However, parameter control problem is still crucial to the performance of algorithms. In this paper, we propose a self-adaptive parameter control method based on entropy and security market line, fully considering the characteristics scheduling problems. It consists of two key parts: locus-entropy strategy and parameter-control strategy. Firstly, the entropy on each genetic locus is calculated to accurately evaluate the population status of scheduling algorithms. Then, a parameter-control strategy based on the conception of security market line is proposed to address the issue that the nature of multipeak in scheduling problems makes algorithms fall into local optimal solutions. The strategy maintains the solutions of good quality and eliminate the solutions of poor quality by using locus-entropy as feedback. Through our method, the balance between exploitation and exploration is kept in algorithms to perform well in scheduling problems with different dimensions and characteristics. These strategies are tightly linked to adjust parameters adaptively without introducing new parameters, so that meta-heuristic algorithms equipped with the proposed approach are able to find a better solution. Moreover, our parameter control method is universal for meta-heuristic algorithms. The proposed approach is hybrid with genetic algorithm and particle swarm optimization. The hybrid algorithms are first compared with the standard algorithms in multipeak benchmark functions, then with other variants of the standard algorithms in real-world single and multi-objective scheduling problems. The results demonstrate that the proposed approach is valid for different kinds of algorithms to enhance the performance of solving a variety of scheduling problems.