Modeling performance and information exchange between fishing vessels with artificial neural networks

Abstract

A fishery is simulated in which 20 artificial vessels learn to make decisions through an artificial neural network in order to search for schools of fish among the available fishing grounds. Three scenarios with different degrees of variability, including uncertainty in the searching process, are considered. The simulation model accounts for the main features commonly observed in a purse seine tuna fishery in a time and a space scale. Vessel strategies are chosen by the artificial neural network, on the basis of the following decision criteria: information concerning time searching in a specific area, previous performance in this area, knowledge of the quality of surrounding fishing grounds, presence of other vessels fishing actively and trip length. An analysis of the effects of sharing information between vessels is done and this was compared to individual artificial fishing vessels. In general, a group of fishing vessels show higher performance than individual vessels. A convex performance comparison curve for several group sizes is found in all scenarios considered. The optimum group size differs according to the variability of the artificial world. At bigger group sizes performance decreases, probably due to competition and depletion effects of some fishing grounds.