Applying artificial intelligence to optimize the trawling path and operational parameters for Antarctic krill

Antarctic krill (Euphausia superba) is a vital species in the Southern Ocean ecosystem and have high commercial value. Effective control of trawling activities targeting Antarctic krill is crucial for enhancing fishing efficiency and minimizing energy consumption resulting from inadequate operations. This study involves a reinforcement learning method that enables interaction between the trawl and krill distribution environment, ultimately learning the optimal trawling path for maximizing fishing efficiency. The neural network was utilized to predict operational parameters within the optimal trawling path. The results demonstrate that the optimized path for capturing a greater quantity of krill required more adjustments of the trawl compared to the actual trawl path, which experienced minimal changes. The density of the krill area passed by the trawl mouth was 8.4% higher when following an optimal path compared to the actual path. The neural network, consisting of three layers with 5 neurons in the hidden layer, accurately predicted the warp length and towing speed for the optimal path. The analysis of generalized additive models confirms a high level of agreement (P = 0.69) between the path based on predicted operational parameters using the neural network and the optimized path. Furthermore, this study demonstrates that adhering to an optimal trawling path for Antarctic krill enhances fishing efficiency and promotes the automation of Antarctic krill trawling operations.