Managing supply uncertainty by a formal stochastic programming approach demonstrates how expected profit may be increased due to more robust production plans.
Operational production planning deals with establishing optimal production plans. A production plan determines which production lines to utilize, which products to produce and allocation of resources such as personnel and machines. At the time of planning the supply of raw material for use in production is unknown with respect to volumes, quality and distribution of species.
A production plan which is optimal in case of high catch volumes may turn out to be very costly in case of a low catch scenario. Stochastic programming is an approach to establish plans that maximize expected profit taking such uncertainty into account.
The research explores a stochastic programming approach for handling supply uncertainties in operational production planning. In the problem structuring phase focus has been on identification of decision variables such as content of different execution plans, how many hours to follow each execution plan, how many workers to allocate for the work and whether to pay extra for over time. Next the various uncertainty factors are made operational, i.e., by assigning probabilities for various scenarios. A scenario is a discretization of the supply with an adding probability, for example high catch, high percentage of high catch quality, medium percentage of haddock and Pollock.
To reduce supply uncertainty a statistical prediction model has been developed. Operations planners in the industry have addressed the challenge of efficient planning under uncertainty and how information regarding type of vessels, fishing gear, weather etc may be used for improved catch prediction and reduction in uncertainty. Finally from literature it is well known that treating uncertainty explicitly in optimization problems outperforms an approach where uncertain variables are replaced by their expected values.
To simplify modelling main focus has been on critical activities such as filleting and cutting in the production line. Only processing of cod is considered, thus future models needs to include other species like haddock and Pollock. Rather than considering the flow of raw material in the production line the modelling treats a limited number of so-called production plans. The main decision variable is how many hours per day a given production plan shall applied.
From the case study the stochastic programming approach gives a 70% higher expected profit compared to the situation where uncertainty is ignored. Refining the model by adding more discretization levels in the scenario building gives even better result, but at the cost of more complex models.
The catch prediction model developed is able to explain more than 60% of the variability by weather forecast and type of vessel as explanatory variables.
The approach has identified an approach where the value of more refined uncertainty models is quantified. This enables the developers of production planning tools to include only the most significant uncertainty factors in a manner where expected gain is balanced with the complexity of the model. Further, the catch prediction model may be used to reduce the supply uncertainty.
A limited number of uncertainty factors have been included, and it remains to investigate uncertainty in quality deterioration rate, initial quality distribution of the catch, production resources and demand/selling prices. The current model only deals with a 3 day period, and hence the model should be expanded to a rolling horizon model. Finally a multi-plant production planning model will demonstrate how production could be distributed across plants in an optimal manner.