基于SMO-SVR的飞机舵面损伤故障趋势预测  

Fault prediction for aircraft control surface damage based on SMO-SVR

作　　者：董磊[1] 任章[1] 李清东[1] 

机构地区：[1]北京航空航天大学飞行器控制一体化技术重点实验室,北京100191

出　　处：《北京航空航天大学学报》2012年第10期1300-1305,共6页Journal of Beijing University of Aeronautics and Astronautics

基　　金：国家自然科学基金资助项目(60874117;61101004)

摘　　要：飞机舵面出现损伤时,为了更准确的预测状态参量变化情况,提出了一种改进的序贯最小优化支持向量回归(SMO-SVR,Sequential Minimal Optimization Support VectorRegression)预测方法.采用改进C-C平均方法对多元时间序列进行相空间重构,以确定最优嵌入维数m和延迟时间τd.根据所求m和τd建立加权SVR预测模型,并调整了SMO算法的停机准则.利用区间自适应粒子群算法(IAPSO,Interval Adaptive Particle Swarm Optimization)优化SVR参数,以提高参数优化速度.为了验证改进算法的有效性,针对飞机方向舵损伤故障趋势进行了预测和分析,并与径向基函数神经网络(RBFNN,Radial Basis Function Neural Net-work)方法进行了对比,仿真结果表明SMO-SVR预测模型具有很好的预测能力.In order to predict changes more accurately when the surface of aircraft damaged, an algorithm based on improved sequential minimal optimization support vector regression （SMO-SVR） was presented. This algorithm reconstructed the phase space of multivariate and nonlinear time series using improved C-C average method to determine the embedding dimension m and the delay time Td. Then, a weighted SVR model was built according to m and td, and in which the halt criterion of SMO was modified. The parameters of SVR were optimized by interval adaptive particle swarm optimization （IAPSO） to improve the efficiency of parame- ter optimization. In order to verify the validity of the algorithm, the prediction and analysis of surface damage trend were performed. Comparing with the radial basis function neural network （RBFNN） method, the simula- tion result demonstrates that the improved SMO-SVR prediction model has good predictive ability.

关 键 词：故障趋势预测 支持向量回归 序贯最小优化 舵面损伤 相空间重构 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]