퍼지논리와 강화학습을 적용한 트롤 시스템 제어

Abstract

In this study, a trawl fishing simulator was developed for various fishing industries in response to the demand of educational institutions and fishery-related companies for automated system operations. In education institutions and fishery-related companies, fishing simulators are used to conduct training to acquire indirect experience on fishing. Training with fishing simulators not only decreases time and costs, but also allows training in various environments. Owing to these advantages, the demand for fishing simulators for education and training is increasing. Therefore, in this study, a trawl fishing simulator was developed based on software in which the trawl operation process can be smoothly run in both personal desktop computers and laptops. Also, an optimized analysis technique was developed through which the dynamic behavior generated by the interaction between fishing vessels and gears can be analyzed in the time domain. The developed trawl simulator can be used for education and training on the entire trawl fishing process, from fish scouting to catching. Unlike previously developed fishing simulators, it has no spatial constraints because it does not require many computers or large hardware, and is also easy and simple to use. Therefore, it shall be very useful for educational institutions and fishery-related companies when providing education or training on trawl operations. In trawl operations, controlling the depth of trawl fishing gears is the most important factor that determines whether fish will be successfully caught or not. It is necessary to accurately estimate the location and depth of the target fish school, control the direction and speed of the trawler, and cast the fishing gear at the exact depth of the target fish school. To automate this process, 3D control methods that allow simultaneous control of the trawler direction and depth of the trawl fishing gear are required. Therefore, in this study, fuzzy control rules were applied to the trawl simulator for the automated control of the trawler direction and depth of the trawl fishing gear. The results of 3D control of the trawl fishing gear by applying fuzzy control rules reveal that the depth of the trawl fishing gear was controlled with an error of less than 5%. Therefore, it is expected that the process of controlling the depth of the trawl fishing gear by adjusting the warp length and controlling the direction of the trawler during trawl operations can be automated. Also, if a conversion device is developed that allows the application of the designed fuzzy control rules to sonars, fish finders, steering gears, and fishing winch systems equipped in actual trawlers, then 3D control of trawl fishing gears using fuzzy control rules will be possible in actual trawl operations. During trawl operations, if fishing gears are cast too early or too late, variations occur in the catch rate and fuel consumption. Therefore, the trawl fishing gear must be cast at the right time to not only draw the target fish school accurately but also to maintain fuel consumption at proper levels. In this study, a Deep-Q-Network was applied to the developed trawl simulator to estimate the appropriate casting time of the trawl fishing gear. The appropriate casting time was estimated by performing a total of 5,000 reinforcement learning sessions. The compensation value gradually increased as the number of reinforcement learning sessions increased, and the optimum trawl gear casting time estimated using the compensation value determined after 5,000 reinforcement learning sessions was applied to the trawl simulator. When the trawl fishing gear is cast at the estimated casting time, the location of the target fish school can be trawled more accurately, and the fuel consumption can also be maintained at proper levels. In the future, applying this reinforcement learning algorithm to actual trawl operations is expected to increase the catch rates of inexperienced executive crews by helping them learn the skills of experienced executive crews.