水下机器人神经网络自适应逆控制

Adaptive Inverse Control of Neural Network for Underwater Vehic les

水下环境的复杂性以及自身模型的不确定性,给水下机器人的控制带来很大困难。针对水下机器人的特点和控制方面所存在的问题,提出了基于预测 校正控制策略的水下机器人神经网络自适应逆控制结构及训练算法。通过在线辨识系统的前向模型,估计出系统的Jacobian矩阵,然后采用预报误差法实现控制器的自适应。同时,为了提高系统对于外扰的鲁棒性,在伪线性回归算法的基础上,在评价函数中引入微分项。理论分析和仿真结果表明,与原来的算法相比,微分项的引入改善了系统对于外扰的鲁棒性和动态性能。

It is difficult to control unde rwater vehicles for its high nonlinear,uncertainty dynamic behavior of the vehicles and robust performance to external disturbances and dynamic un certainty of the vehicle. A neural network based adaptive inverse control struct u re and training algorithm for autonomous underwater vehicles is proposed base on the predictor-corrector control strategy. The Jacobian matrix is e stimated through the vehicles' forward NN model.And adaptation of the controller is realized by means of the predictive error method.The differentiation item is introduced to the critic function base on the algorithm proposed by others.The results by the oretical analysis and simulation have shown that the robust and dynamic performa nce of the system is improved compared to the previous learning algorithms.