A Neural Network Based Adaptive Autopilot for Marine Applications

Due to varying dynamics of sea-going vessels with changes in gross tonnage, speed of the vessel and depth of water, the key factor limiting the performance of current autopilot systems using PID/PD controllers, is the wide range of vessels' dynamical behavior. Previous research proposes adaptive controllers to overcome these difficulties, but such systems can also suffer from disadvantages such as potential instabilities. This paper investigates the application of artificial neural networks to automatic yaw control of mine sweepers at various speeds, where the practical issues like speed of response relevant to this particular class of ship are carefully considered. The proposed networks are trained offline to capture the controllers' dynamics and use it in the control loop thus incorporating the properties of a series of conventional PD controllers designed at different forward speeds and hence improves the vessel's automatic steering performance under a variety of operational conditions