基于混合学习算法的水下机器人神经网络辨识

Neural network identification of underwater robot based on hybrid learning algorithm

借鉴Elman和Jordan神经网络的特点,构造了一种新的动态神经网络.该网络能对隐含层的历史进行状态记忆,实时调整过去的信号对现在值的影响,并且增加了输出层节点的反馈以增强神经网络的信号处理能力.将基于遗传算法(GA)和误差反传算法(BP)的混合学习算法用于神经网络权值的修改,既可提高收敛速度又能避免陷于局部极小值.最后,将改进的神经网络应用于水下机器人动力学模型辨识,仿真结果表明,基于混合学习算法的神经网络提高了学习的收敛速度和辨识精度.

A new dynamic neural network is constructed by borrowing ideas from Elman and Jordan neural networks. The new network can remember the history state of hidden layer and tune the effect of the past signal to the current value real-timely And in the presented network, the feedback of output layer nodes is increased to enhance the ability of handling signals. A hybrid learning algorithm based on Genetic algorithm （GA） and error back propagation algorithm （BP） is used to tune the weight values of the network, which can accelerate the rate of convergence and avoid getting into local extremum. Finally, the improved neural network is utilized to identify the A UV hydrodynamic model, and the simulation results show that the neural network based on hybrid learning algorithm improves the learning rapidity of convergence and identification precision.