基于神经网络的飞参数据特征选择方法

针对飞参数据中存在的大量冗余和不相关,提出了一种基于神经网络的飞参数据特征选择方法.为克服传统算法收敛速度慢、易陷入局部极小等缺陷,神经网络的训练采用粒子群优化算法和Levenberg-Marquardt优化算法相结合的方式.神经网络训练结束后,先利用网络权值信息对飞参数据特征的相对重要度进行排序,然后根据重要度次序对飞参数据特征进行选择.实验结果表明该方法能快速有效地删除冗余飞参数据特征,同时提高网络的泛化能力.

Feature selection method of flight data based on neural network

Aimed at the redundant and irrelevant features of flight data, a new feature selection method of flight data based on neural network is proposed. In order to overcome the disadvantages of traditional algorithms such as slow convergence and local minimum, the neural network is trained by an improving method synthesizing particle swarm optimization algorithm and Levenberg-Marquardt optimization algorithm. When the neural network is trained well, feature importance of flight data is ranked using network weights, and then feature selection of flight data is carried out based on the rank of feature importance. The experimental results show that the new method remove redundant feature selection effectively, and improve the generalization ability of neural network as well.