Predicting weight dispersion in seabass aquaculture using Discrete Event System simulation and Machine Learning modeling

Abstract

Marine aquaculture, particularly in the Mediterranean region, faces the challenge of minimizing growth dispersion, which has a direct impact on the production cycle, market value and sustainability of the sector. Conventional grading methods are resource intensive and potentially detrimental to fish health. The current study presented an innovative approach in predicting fish weight dispersion in European seabass (Dicentrarchus labrax) aquaculture. Seabass is one of the two major fish species cultivated on the Mediterranean coast, with a fattening cycle of 18–24 months. During this period, several grading operations are carried out to minimize growth dispersion. The intricate feed-fish-water system, characterized by complex interactions among feeding regimes, fish behavior, individual metabolism and environmental factors, is the focus of the study. The comprehensive, five-step methodology addresses this complexity. The process begins with a Discrete Event System (DES) model that simulates the feed-fish-water dynamics, taking into account individual fish metabolism. This is followed by the development of a surrogate machine learning (ML) regressor model, which is trained on DES simulation data to efficiently predict growth distribution. The model is then calibrated and customized for specific fish stocks and production tanks. The preliminary results from 21 tanks in two trials with European seabass (D. labrax) showed the effectiveness of the method. The results from the simulation models achieved a R2 of 99.9 % and a Mean Absolute Percentage Error (MAPE) of 1.1 % for the prediction of mean final weight and a R2 of 90.3 % with a MAPE of 8.1 % for the standard deviation of final weight. In summary, this study represents a significant advance in the planning and management of seabass aquaculture. Given the lack of effective prediction tools in the aquaculture industry, the proposed methodology has the potential to reduce risks and inefficiencies, thus possibly optimizing aquaculture practices by increasing sustainability and profitability.

Keywords

• Aquaculture
• Fish weight dispersion
• Discrete event systems
• Machine learning
• European seabass
• Fish growth modeling