In the Simulated Annealing (SA) algorithm, the Metropolis algorithm is applied to generate a sequence of solutions in the search space, known as the Markov chain. Usually, the algorithms employ the same Markov Chain Length (MCL) in the Metropolis cycle for each temperature.
However, SA can use adaptive methods to compute the MCL.
This work aims to analyze the effect of using different MCL strategies in SA behavior. This experimentation considers the Water Distribution Network Design (WDND) problem, a multimodal and NP-hard problem interesting to optimize. The results indicate that the use of adaptive MCL strategies improves the solution quality versus the static one.