Mechanistic model based optimization of feeding practices in aquaculture

Abstract

Fish feed accounts for more than 50% of total production cost in intensive aquaculture. Feeding fish with low quality feed or adopting inappropriate feeding strategies causes not only food waste and consequent loss of income but also lead to water pollution. The aim of this study was to develop a mechanistic model based optimization method to determine aquaculture feeding programs. In particular, we integrate a fish weight prediction model and a requirement analysis model to establish an optimization method for designing balanced and sustainable feed formulations and effective feeding programs. The optimization strategy is necessary to maximise the fish weight at harvest, while constraints include specific feed requirements and fish growth characteristics. The optimization strategy is re-solved with new available fish weight measurement by using the error between measurement and model prediction to adjust the requirement analysis model and update feeding amount decision. The mechanistic models are parameterised using the existing nutritional data on gilthead seabream (Sparus aurata) to demonstrate the usefulness of proposed method. The simulation results show that the proposed approach can significantly improve aquaculture production. This particular simulation study reveals that when “Only prediction” method is considered as benchmark, the average improvement in fish weight of proposed method would be 13.25% when fish weight is measured once per four weeks (mimicking manual sampling practice), and 38.43% when daily measurement of fish weight is possible (e.g. through automatic image-based methods). Furthermore, if feed composition (460 g protein.kg feed−1 ; 18.9 MJ kg feed−1 ) is adjusted, the average improvement of proposed method could reach 46.85%. Compared with traditional feeding methods, the improvement of proposed method could reach 36.36% of the final fish weight at harvest. Further studies will consider improving the quality of feed plus executing more appropriate mathematical prediction models to optimize production performance.