ARTIFICIAL INTELLIGENCE AND HIGH-PERFORMANCE COMPUTING FOR DECISION MAKING IN AQUACULTURE

Introduction

 Aquaculture has become one of the fastest-growing industries in the world due to the constant increase in demand, limitations in fishery supply, and innovations in the production process. However, decision-making and planning processes remain particularly complex as they are affected by many technical, biological, environmental, and socioeconomic factors.  Therefore, the aquaculture industry is still considered a high-risk and complex industry, in which decisions made throughout the production process are not based entirely on planning and control systems.

 In this context, industries would benefit from technological innovation in management and decision-making support systems aiming to maximize efficiency and minimize the  environmental impact and risk s associated with  this activity. In this regard, the main objective of this work is the development and evaluation of new optimization methodologies based on the use of Artificial Intelligence techniques and high-performance computing, to support aquaculture producers in decision-making processes .  Thus, t he capacity  and effectiveness of these techniques to optimize the results of marine farms , from both economic and environmental points of view, will be evaluated.

Materials and Methods

 This study follows a two-phase process to achieve the aforementioned objective:

•  First, a methodology that integrates a multicriteria model and a Particle Swarm Optimization (PSO) technique, proposed by Luna et al. (2020), is applied to find the optimal production strategy in a farm with multiple cages, batches, feeding alternatives, and products. In this case, t his method is focused on improving the results of aquaculture companies from a multicriteria point of view (Figure 1), with low computational cost, addressing decisions regarding stocking and harvesting and the synchronous selection of feed and fry for a large number of cages.
• The optimization process described above would entail a very high computing time  due to not  only the fact that  companies  tend to have a n increasing number of production units (farms) but also because of the large number of decision criteria included. For that reason, parallelization techniques in distributed systems have been used, based on the techniques published by (Ibáñez et al., 2023), to  improve the model computational cost, in terms of  performance and energy consumption.

 The integration of these techniques is  especially important since one of the keys to these methods, and what would allow them to be scaled to different production systems, is their ability to find good solutions in a reduced time, even under complex conditions (Blum and Roli, 2003). To this end, the efficiency of the previous design will be compared to new parallel developments executed with 60 and 80 cores, respectively.

Results

Th at  methodology has been tested for  different scenarios of  sea bream marine farming with respect to location, sea temperature, and scheduling dates. The information used has been collected from primary sources, such as oceanographic buoys or Spanish market prices, and secondary sources of information, as described by Luna et al. (2020).

On the one hand, the optimization phase has proven to be useful for aquaculture farms as it could achieve  an improvement in the aggregated results for the criteria of Figure 1 between 20% and 50%.  On the other hand,  it proves the  importance of high-performance computing in th ese scenarios as they face exponential growth in the computational cost with increasing complexity, either due to the increase in the number of production units or the inclusion of constraints in the process.

A s can be seen in Figure 2, parallelization techniques in distributed memory systems (with 60 and 80 cores) have shown great ability to significantly reduce execution time. The use of these techniques has allowed the optimization of 80 cages in a time of 12 minutes or even 200 in just over 20 minutes. Additionally, the problem of searching for feasible solutions has also been solved due to the possibility of improving optimization parameters without experiencing a significant impact on computing time.

Conclusion

 Decision-making processes are particularly complex in the aquaculture industry. But, as the results of the present study show, decision-support techniques and systems could be the perfect solution to optimize the results of aquaculture companies, from both economic and environmental side.

This is in line with previously published studies regarding the application in the aquaculture industry of decision support systems (Cobo et al., 2019) or multicriteria decision-making techniques (Vergara-Solana et al., 2019). However, this study shows that  parallelization techniques in distributed memory systems allow decision-makers to achieve that in sufficiently low computing times, enabling its application to large companies or conglomerates of companies. Furthermore, these results open the door to the application of these techniques in aquaculture companies with different production processes, some with hundreds of production units, and to decisions and scenarios of greater technical complexity.

References

Cobo , A., Llorente, I., Luna, L., & Luna, M. (2019). A decision support system for fish farming using particle swarm optimization. Computers and Electronics in Agriculture 161, 121-130.

Blum , C., & Roli, A. 2003. Metaheuristics in combinatorial optimization. ACM Computing Surveys, 35(3), 268-308.

 Ibáñez, M., Luna, M., Bosque, J. L., & Beivide , R..  (2023). Parallelisation of decision-making techniques in aquaculture enterprises. The Journal of Supercomputing.

 Luna, M., Llorente , I., & Cobo, A. 2020. Aquaculture production optimisation in multi-cage farms subject to commercial and operational constraints. Biosystems Engineering, 196, 29–45.

 Vergara-Solana, F., Araneda, M. E., & Ponce-Díaz, G.. (2019). Opportunities for strengthening aquaculture industry through multicriteria decision-making. Reviews in Aquaculture, 11(1), 105–118.