Multi-stage safe ship control with neural state constraints

The paper describes an application of the dynamic programming method to determine own ship's safe trajectory during the passing of other ships encountered. A dynamic model of the process with kinematic constraints of state determined by a three-layer artificial neural network has been used for the synthesis of control. Non-linear activation functions in the first and second layers may be characterised by a tangent curve while the output layer is of a sigmoidal nature. Neural Network Toolbox from the Matlab package has been used to model the network. The learning process used an algorithm of a backward propagation of the error with adaptively selected learning step. The considerations have been illustrated with example of computer simulation the algorithm of determination the safe ship's trajectory in situation of passing 18 ships encountered, recorded on the ship's screen radar in real navigational situation in Skagerrak Strait.