<p>An automatic trawl winch control integrated with warp-tension monitoring and a machine learning model was designed with the aim of shifting to an environmentally friendly operation technology. Towing experiments with reduced-scale bottom otter trawl gears were conducted to gather time-series data on warp tension, which were used to construct a model for identifying towing states, including pelagic, semipelagic, bottom, and overturn trawls. This model, based on a three-layer fully coupled neural network model, was then modified by a genetic algorithm and subsequently incorporated into the automatic control system. The results show that, although there were no clear thresholds of the warp tension available to easily recognize the towing states of the otter trawl, the modified towing-state identification model achieved an accuracy of about 80% recall. Moreover, the automatic control system equipped with the towing state identification model successfully managed transitions between pelagic or bottom trawl to semipelagic operations, and it was confirmed to have an identification accuracy of over 80%.</p>