Distributionally Robust Chance-Constrained Bin Packing

Chance-constrained bin packing problem allocates a set of items into bins and, for each bin, bounds the probability that the total weight of packed items exceeds the bin's capacity. Different from the stochastic programming approaches relying on full distributional information of the random item weights, we assume that only the information of the mean and covariance matrix is available. Accordingly, we consider distributionally robust chance-constrained bin packing (DCBP) models. Using two types of ambiguity sets, we equivalently reformulate the DCBP models as 0-1 second-order cone (SOC) programs. Furthermore, we exploit the submodularity of the 0-1 SOC constraints under special and general covariance matrices, and derive extended polymatroid inequalities to strengthen the 0-1 SOC formulations. We then incorporate these valid inequalities in a branch-and-cut algorithm for efficiently solving the DCBP models. Finally, we demonstrate the computational efficacy of our approaches and performance of DCBP solutions on test instances with diverse problem sizes, parameters, and item weight uncertainty.