大射电望远镜舱索系统的神经网络预测模型的建立

Establishment of NN Prediction Model for Cable-cabin Movement System of the Large Radio Telescope

由大跨度柔性悬索拖动馈源舱运动来实现其精确定位的新型大射电望远镜(LT)具有变结构、非线性、大滞后、多输入多输出的特点,传统的建模方法已很难建立起其精确的数学模型,这里提出了采用BP神经网络进行辨识建模的思想,考虑到基于标准BP算法的神经网络收敛速度慢、易于陷入局部极小值的不足,提出了基于数值优化的L M(Levenberg-Marquardt)训练算法。应用Matlab6.x中的神经网络工具箱实现了系统的仿真,实验结果表明,采用这种方法可成功地建立舱索系统模型,无论其学习能力还是泛化能力都得到了很好的效果,且其收敛速度大大提高。

Pulled by a flexible long cable, the cable-cabin system of the new large radio telescope(LT) has characteristics of variable structure, nonelinearity, time delay and MIMO. It was difficult to establish a precise model by the traditional methods, so a BP neural network was utilized to identify the system model. Considering the disadvantages of slow convergence velocity and the tendence to converge to a local minimum, a L-M(Levenberg-Marquardt) training algorithm was suggested. Further more, simulation experiments were done using matlab6.x neural network toolbox. The results show that the prediction model has good learning ability and generalization ability, and its convergence velocity is improved considerably.