A neural-network approach to the control of surface ships

✍ Scribed by R. Burns; R. Richter

• Publisher: Elsevier Science
• Year: 1996
• Tongue: English
• Weight: 512 KB
• Volume: 4
• Category: Article
• ISSN: 0967-0661
• DOI: 10.1016/0967-0661(96)00019-6

✦ Synopsis

Conventional ship autopilots are based on proportional, integral and derivative (PID) algorithms, and are generally set to work under specific conditions. Changes in either the vessel's handling characteristics or environmental conditions means that the system is not working at its optimal point. This paper explores the possibility of developing two neural network autopilots based on training data derived from: a) a small vessel operating in a range of sea states, using differently tuned PID controllers for each sea state.

b) an optimal guidance system for a large ship sailing in calm water at varying forward speeds.

It is demonstrated that with the small vessel, a single neural network can cope with a range of sea states without the need for re-tuning. In the case of the large vessel, the trained network performed in a slightly sub-optimal manner -but had the advantage that it was not necessary to re-compute controller parameters at different forward speeds.