基于改进的神经网络的船舶碰撞危险度的模型

针对船舶碰撞危险度具有模糊性、不确定性等特点,依据模糊理论方法建立的船舶碰撞危险度的数学模型,直接采用来船航速、来船航向、来船对本船的相对舷角和来船对本船距离作为神经网络的输入,采用Levenberg-Mrquardt优化算法这种改进的BP神经网络进行训练和仿真,并与标准BP算法和动量BP算法进行比较,发现经过改进的网络求得碰撞危险度比标准BP算法和动量BP算法具有更好的效果,网络能够更有效收敛,大大提高了网络的收敛速度和泛化能力。

An Improved Neural Network Based Model of Collision Risk Index

For the characteristics of illegibility and uncertainty of vessel collision risk index a mathematical model of vessel collision risk index is set up based on fuzzy theory,which takes the coming vessel's velocity and course,the relative course and distance between the coming vessel and own vessel as inputs of network and adopts Levenberg-Mrquardt algorithm to train and simulate the network..Compared to traditional BP networks,this improved neural network has better effectiveness for prediction of collision risk index,which can be used conveniently in ship automatic collision avoidance systems.