Operational Support in Fish Farming through Case-Based Reasoning

Farmed fish is the third biggest export in Norway (around NOK 30 billion/e3.82 billion/US$ 5.44 billion in 2010), and large fish farms have biomass worth around NOK 150 million/e19.38 million/US$ 26.72 million. Several processes are automated (e.g. the feeding system), and sensory logging systems are becoming ubiquitous. Still, the key to successful management of a site is the operational knowledge possessed by the fish farmers. In most cases, this information is not stored formally. To capture, store and reuse this knowledge in a more systematic way is called for. We present a system that employs case-based reasoning (CBR) for such knowledge management, combined with sensor data and numerical models. The CBR system will ultimately be the core part of a decision support for regional managers surveying fish farming sites. Data is acquired from multiple fish farms, spanning several years. We present recent results in testing how well the CBR system finds similar cases. An important part of this test is the evaluation of three different methods for case retrieval (kNN, linear programming for setting feature weights, Echo State Network).