Optimization models and methods for production planning and ship scheduling in the pulp industry

Abstract: To use optimization models and methods is an increasingly acknowledged element of supply chain management for large industries. Concurrently with the increasing global market competition and modern business policies, an already effective organization needs to find new possibilities to improve their planning procedures and resource utilization. To incorporate the planning in a more comprehensive context, in an overall supply chain, provides new possibilities to efficiently manage several planning steps than if considered individually.This thesis considers production planning and ship scheduling problems at Södra Cell, a large Scandinavian pulp manufacturer. The main purpose of the thesis is to suggest optimization models and methods that can support production and distribution planners to objectively consider the supply chain impact in the short-term decision making process. The production planning and the ship scheduling problems are approached separately. Optimization models are formulated for both problems, and specially adapted solution methods are developed that account the characteristics of the problems.The thesis consists of three research papers. In the first paper two mixed integer models (MIPs) are developed for the production planning problem. The first model has binary variables for every possible production plan and a column generation approach with a constraint branching heuristic is used to solve the otherwise very large model. The second model has the production choices formulated in constraints and is solved with a standard branch and bound search. The results are compared with manual planning and potential large savings are indicated.The second and the third paper both consider the pulp distribution problem where ship routing and scheduling decisions have a key role. In the second paper the problem is solved with a rolling time horizon method. An MIP model is solved repeatedly over short ranges of time until eventually the whole planning period is covered. The third paper presents a genetic algorithm with integrated linear programming models for the distribution problem. The solutions from the different solution strategies for the distribution problem are mutually compared in the third paper.