Binary Decision Diagrams (BDDs) have become the state-of-art data structures in numerous fields, wherein, small-sized BDDs are required to reduce the companion cost. Since BDD size is sensitive to the order of variables for the Boolean function in use, evolutionary algorithms
have been extensively exploited to solve the BDD reordering problem which is provably an NP-hard problem. However, getting trapped in a local minima is more likely as the population gets homogeneous during the evolution of the algorithm. This paper compares various diversity
measures correlated to BDDs and studies the impact of different crossover and mutations used in the Genetic Algorithm (GA) on those diversity measures. Thereafter, we propose an enhanced GA-based reordering algorithm for BDD, wherein, the chosen diversity measure is calculated per generation to steer the application of proper variation operators, in such a way that an acceptable level of equilibrium between exploration and exploitation is preserved. Finally, we utilize a new heuristic Cyclic Crossover (HCX) that further improves the performance of the proposed algorithm following the fitness value. Experimental results on public benchmarks show that our proposed algorithm outperforms other algorithms from the literature when switching between HCX and swap mutation operators is made following the measured diversity metric.

Title

Applied Soft Computing

Publisher

Elsevier

Publisher Country

Netherlands

Indexing

Scopus

Impact Factor

8.7

Publication Type

Both (Printed and Online)

Volume

--

Year

2024

Pages

--