Automatic steering of ships using neural networks

Abstract

SUMMARYShip steering control system design presents challenges because the dynamic properties of the vessel itself vary signi"cantly. The use of an arti"cial neural network as a controller which incorporates the properties of a series of conventional controllers designed for di!erent operating conditions could provide an alternative to adaptive control or gain scheduling in this application. Local model network methods could also provide a basis for e$cient modelling of the vessel over a range of operating conditions. The paper describes an investigation of radial basis function networks for ship steering control and of local model networks for representation of ship dynamics. Performance is demonstrated by a series of simulation studies.

Control for Ship Course-Keeping Using Optimized Support Vector Machines

“…Conventional control strategies like PID control can no longer satisfy the requirement of navigation, guidance and control of ships. During recent decades, many advanced control schemes have been developed and successfully applied to ship motion control, including sliding mode variable structure control [2,3]; parameter adaptive control [4,5]; H-infinity robust control [6,7]; neural-network control [8,9]; fuzzy control [10,11]; neuro-fuzzy control [12,13]; line-of-sight based model control [14,15], etc.…”

Design of Cooperative Output Regulators for Heterogeneous Uncertain Nonlinear Multiagent Systems

“…2. The system parameters for the ships are T 1 = 21, K 1 = 0.23, α 1 = 0.3 used in [34], T 2 = 31, K 2 = 0.5, α 2 = 0.4 used in [35], T 3 = −1252.3, K 3 = 0.084, α 3 = 1, and T 4 = −249.9, K 4 = 0.0119, α 4 = 1 used in [36]. The Nussbaum gains used for the ships are N(k 26) is σ = π/300.…”

“…, r N ), r i > 0, for 1 ≤ i ≤ N, such that RH+H T R ≥ λ 0 I. Thus, the first term in (35) can be expressed as…”

“…for some constants ϕ α i , ϕ β i > 0. Under this condition, for all l > 0, the second term in (35) can be bounded by…”

Survey of Deep Learning for Autonomous Surface Vehicles in the Marine Environment

“…• Path Following: Without dependency on prior knowledge of dynamic modeling, DNN [166]- [170] and DRL [14], [19], [175] are able to generate the control signal directly. DRL can learn from the interactions between the agent and environment to find the best policy without knowing any information in advance.…”