Exploiting data to support operations of EXPOSED aquaculture installations

PhD-candidate: Bjørn Magnus Mathisen

Supervisors: Professor Agnar Aamodt, Professor Helge Langseth, Professor Kerstin Bach

Duration: Q3 2015 – Q2 2020

Aquaculture is an important industry in Norway, and has always maintained a strong focus on safety for human and non-human resources as well as product quality. This industry is now seeking new sites for its fish farms. However many of these sites are remote and are subject to harsh sea and weather conditions.

This PhD project aims to contribute to safe and sustainable production at these site through utilization of human experience combined with collected data (both onsite data and other data sources) for monitoring and operational decision support. The project will study whether automation through machine learning and decision support systems can improve safety, production quality and effiency on the new exposed sites. To answer these questions, we need to:

  1. Identify core targets for automation and decision support in collaboration with domain experts.
  2. Gather and validate relevant data.
  3. Identify and develop machine learning method(s) for decision support
  4. Measure performance of proposed method(s)
  5. Validate method(s) with real world tests.

Current work: I am currently in the final phases of finishing work on a second paper where we are working on methods for measuring similarity or distance between two data points. These similarity metric will be learned from recorded data using deep neural networks (See Figure 1). This will be tested on data provided from project P8 and Anteo AS. We will compare operational situations for exposed aquaculture: E.g. “Is todays delivery of fish feed similar to any previous such operation at this location which needed special intervention?”

Figure 1: Architecture for a neural network designed for measuring similarity between datapoints

List of published papers:

  • Bjørn Magnus Mathisen, Agnar Aamodt, and Helge Langseth. “Data driven case base construction for prediction of success of marine operations.” CEUR Workshop Proceedings, 2017.