基于Elman模型的动态障碍物运动状态估计

Dynamic Obstacle Motion State Estimation based on Elman Model

提出一种基于机器学习的预测模型,估计动态障碍物的运动状态,主要用于无人艇对海上动态障碍物进行自主避碰。系统在神经网络算法的基础上进行改良,设计出神经网络预测控制结构,然后根据控制结构中的环节逐步实现设计。第一步建立神经网络预测模型,基于人工神经通过加入内部反馈信号来描述非线性动力学问题;第二步建立反馈矫正模型及参考轨迹;第三步设计滚动优化算法;最后对算法进行仿真。实验结果表明,经度和纬度的误差都在0.8 m以内,验证了基于Elman算法预测模型的可行性,为后续研究奠定了基础。

A prediction model based on machine learning is proposed to estimate the motion state of dynamic obstacles,which is mainly used for autonomous collision avoidance by unmanned boats against dynamic obstacles at sea.The system is improved on the basis of Elman.Firstly,the neural network predictive control structure is designed,and then the design is realized step by step according to the links in the control structure.The first step is to build a neural network prediction model,which is based on the artificial neural network.The second step is to establish the feedback correction model and reference trajectory.The third step is to design the rolling optimization algorithm.Finally,the algorithm is simulated.The experimental results indicate that the error of longitude and latitude is within 0.8m,which verify the feasibility of the Elman algorithm-based prediction model and lay a foundation for subsequent research.