Abstract: Back Propagation(BP) neural network can be used to forecast Gyroscopic drift. But BP algorithm inclines to fall into local extremum and its training speed is very slow. To overcome the limitation, this paper presents a novel training method based on immune algorithm. The sample output is regarded as the antigen and the weight matrix as antibody. Satisfied antibody can be found after cloning, mutating and restraining. The trained immune neural network is used to forecast gyroscopic drift. Simulation experiments reveal that the algorithm is effective, and the model leads to a precise result in gyroscopic drift forecasting.

Key words: immune algorithm, Neural Network(NN), Back Propagation(BP) algorithm, gyroscopic drift, forecasting

摘要： BP神经网络可用于预测陀螺飘移误差，但容易陷入局部极值，训练速度很慢。针对上述缺点，该文提出了一种基于免疫算法的神经网络，以样本输出为抗原、神经网络权值矩阵为抗体，通过克隆、变异、抑制等步骤找到最优抗体，将最优抗体用于陀螺仪漂移预测。仿真试验显示，免疫训练算法能有效优化网络权值，基于该模型的漂移预测精度较高。

关键词: 免疫算法, 神经网络, BP算法, 陀螺仪漂移, 预测

CLC Number: 

• TP18