Abstract
In this article we propose an accelerated scenario updating heuristic to solve a real-world large-scale multi-stage stochastic mixed-integer model. Motivated from challenges we are facing in sawmills, the model corresponds to multi-period, multi-product production planning including setup constraints, with random yield and demand. While addressing real-life size instances of the problem, the resulting large-scale multi-stage stochastic mixed-integer model cannot be solved by commercial optimisation packages. Moreover, as the production planning model is a mixed-integer program without any special structure, developing decomposition and cutting plane algorithms to obtain good approximate solutions in reasonable time is not straightforward. We propose a successive approximation heuristic inspired from scenario updating heuristic which has been proposed for multi-stage stochastic models with partial recourse. The latter solves the problem by considering only a subset of scenarios which is updated at each iteration. We modify the scenario updating heuristic in two directions: (1) we modify the bounds on the optimal solution so as to make them valid for multi-stage stochastic models with full recourse, and (2) we propose a new scenario selection rule so as to increase the rate of convergence and the quality of solution. Computational experiments for a real-world large-scale sawmill production planning model verify the effectiveness of the proposed solution strategy in finding quickly good approximate solutions.
Acknowledgement
The authors would like to acknowledge the financial support provided by the Forest E-business Research Consortium (FOR@C).