基于飞参数据的飞机操纵系统故障评估方法

Fault Diagnosis of Aircraft Control System Based on Flight Data

摘要:

针对某型飞机的操纵系统故障评估问题，提出了一种基于飞参数据建立的差分进化极限学习机(DE-ELM)算法。该算法融合了差分进化(DE)和极限学习机(ELM)两种算法，通过对飞参数据进行训练，构建了飞机操纵系统的黑箱模型。由于极限学习机(ELM)的输入权值以及隐含层阈值是随机产生的，所以ELM的随机性较大，稳定性不高，故利用寻优能力较强的DE对ELM输入权值和隐含层阈值进行寻优，从而实现ELM的结构优化，提升ELM的稳定性和鲁棒性。仿真结果表明，DE-ELM算法的决定系数达到了97.6%，其均方误差相比于BP神经网络降低了约79%，相比于单纯的ELM降低了64%。所以说该法可以有效提高精确度，同时具有更加良好的泛化性能。

Abstract:

A Differential Evolution Extreme Learning Machine (DE-ELM) algorithm based on flight data is proposed to solve the problem of fault evaluation of an aircraft’s control system. The algorithm combines Differential Evolution (DE) and Extreme Learning Machine (ELM), by training the flight data, a black box model of aircraft control system is constructed. Because the input weights and hidden layer thresholds of the ELM are generated randomly, the randomness of ELM is large and the stability of ELM is not high. Therefore, the input weights and hidden later thresholds of ELM are optimized by DE, which has strong optimization ability, so that the structure of ELM can be optimized and the stability and robustness of ELM can also be improved. The simulation results show that the decisive coefficient of DE-ELM reaches 97.6%, and its mean square error is reduced by 79% compared with BP neural network and 64% compared with ELM. Therefore, this method can effectively improve the accuracy and has better generalization performance.