Cross-Entropy as an Optimization Method for Bridge Condition Transition Probability Determination

Abstract

Much research has been carried out in recent years as to the most appropriate manner in which to obtain suitable transition probabilities for use in Markovian-based infrastructural deterioration models. Although many optimization algorithms have been applied in the process, a review of the literature has shown that the methods in practice may benefit from the use of more sophisticated search algorithms. Consequently, this paper focuses on the applicability of cross-entropy, a previously developed iterative method, as a robust, efficient solver for use in the derivation of these probabilities. An advantage of this method is that through the use of a smoothing parameter, it tends to avoid becoming stuck in the bound constraints of the problem. This value is usually taken to be arbitrarily between 0.6 and 0.9 but was included in this paper as another variable to be optimized, thereby adding to the accuracy of the method. Two different objective functions were compared and the use of stationary versus nonstationary Markov chains was also investigated.