Variable neighborhood descent heuristic for solving reverse logistics multi-item dynamic lot-sizing problems

A hybrid CPU-GPU parallelization scheme of variable neighborhood search for inventory optimization problems

“…The experimental computational study compared the performances of i) the serial implementation of VND presented in [7], ii) an OpenMP implementation of GVNS, iii) an OpenACC implementation of GVNS, and finally iv) a hybrid OpenMP -OpenACC implementation of GVNS. The results presented in [7] were based on a VND and due to its serial implementation no shaking could be succesfully completed in a short amount of time. However, the new parallel implementations provided us the ability to successfully complete the shaking phase in the given time period and thus we now present more complete VNS schemes.…”

“…Each implementation was tested using a recent set of large benchmark instances presented in [7] with 300 products and 52 periods, and executed for 60 secs. The results of the comparative computational study are depicted in In this paper a new hybrid CPU-GPU GVNS metaheuristic algorithm based on OpenMP and OpenACC programming languages was proposed.…”

“…Thus, one can find either parallel VNS heuristic implementations using either CPU parallelization schemes (MPI VNS [2], OpenMP / threads VNS [3,5]), or even hybrid CPU-GPU VNS parallelization schemes [1]. This paper considers a hybrid CPU-GPU parallelization of a variable neighborhood search method that was recently presented in [6,7] for the solution of the uncapacitated multi-item economic lot-sizing problem with remanufacturing. The remainder of this note is organized as follows.…”

A theoretical and computational analysis of lot-sizing in remanufacturing with separate setups

“…As we have already noted in the computational results, exact separation of valid inequalities is often time consuming, and there is a need to develop customized methods for computational efficiency. One immediate direction is to develop heuristic separation algorithms in order to generate only a small number of effective inequalities, and then integrate these with an overall successful heuristic method, such as the variable descent algorithm of Sifaleras and Konstantaras (2017). Such integrated methods often benefit from the strengths of both exact and heuristic methods, e.g., valid inequalities directing the heuristic search in more promising regions while computational times not being hindered thanks to heuristic components.…”

“…As proven by Retel Helmrich et al (2014), the problem with separate setups is − NP hard if even all costs are time invariant, which provides further motivation for the problem we study. The recent papers of Cunha et al (2017); Sifaleras and Konstantaras (2017) propose innovative solution methods for remanufacturing problems with multiple items, and the study of Sahling (2013) proposes a column generation approach for the case with capacities. Finally, it is noteworthy to remark the recent studies of Agra et al (2016) and Attila et al (2017) proposing robust optimization approaches in order to address some inherent uncertainties in these problems.…”