Rollout-Based Routing Strategies with Embedded Prediction: A Fish Trawling Application

Abstract

In an effort to replenish stocks after years of overfishing, the European Union has been implementing policies to restrict the catch in all fisheries including in the Baltic Sea. These regulations incentivize fishermen to seek operational efficiency to counter lower catch limits. In this study, we address the fish trawling problem as a stochastic dynamic orienteering problem and propose rollout-based routing policies. Using a prescriptive analytic approach, we propose several prediction models and combine them with dynamic routing policies, the combination of which forms a prescriptive model. We evaluate and validate the effectiveness of these prescriptive models using a Baltic dataset on the fish species cod (Gadus). We propose a spatiotemporal cross-validation procedure to fairly assess different prescriptive models. Our findings show that reoptimization-based rollout strategies coupled with simple prediction methods such as nearest neighbors perform better than prescriptive models that use complex spatiotemporal smoothing techniques.