改进递阶自适应遗传神经网络在动力调谐陀螺随机漂移建模预测中的应用

Application of Neural Network Based on Improved Hierarchical Genetic Algorithms to the Simulation and Prediction of Dynamically Tuned Gyroscope Drift

针对惯性导航系统中陀螺仪的随机漂移特性,一般采用神经网络建模来对陀螺性能进行预测。但BP网络在网络结构设计及权值、阈值训练上存在很多不足,应用递阶自适应遗传算法来同步优化网络的结构和权值、阈值,使神经网络得到整体的优化。预测结果表明,这种预测方法对于陀螺漂移建模及稳定性预测是可信的。

Focused on the gyroscopes drift characteristic, a new model was adopted for the drift of gyroscopes based on neural networks theory. But common BP algorithm has a 10t of shortcomings in the networks structure, weightand threshold value training method. In this paper, the hierarchical GA was used for optimize BP neural network structure, weight and threshold at the same time. The results show this forecasting method is effective on modeling and forecasting for gyroscopic drift.