Optimal adaptive filtering using finite banks of Kalman filters applied to automatic steering of ships

The problem of minimizing energy loss in adaptive automatic steering of ships is considered. The paper presents the method chosen for the adaptive state estimation phase of the overall control problem. A maximum likelihood technique is employed for adaptively selecting the best Kalman filter from a bank of filters to estimate the ship/steering and seaway model states. The bank consists of previously designed and tested filters for different seaway conditions. The use of this methodology enables the state and parameter estimation problems to be treated separately. The overall adaptive algorithm is structurally simpler than an algorithm which performs these tasks in a combined manner, such as in extended Kalman filtering. With this approach no problem of filter divergence can occur, which is a factor that improves the overall reliability of the autopilot design. Results of simulation studies presented show that the method works very efficiently in selecting the filter gains for the controller to achieve the best performance with respect to minimization of steering related propulsion losses under different seaway conditions.