基于结构最优化RBF神经网络的润滑油金属含量预测  

Prediction for Lubricating Oil Metal Content Based on Structure Optimum RBF Neural Network

作　　者：石宏[1] 张帅[1] 李昂[1] 

机构地区：[1]沈阳航空航天大学航空宇航工程学部,辽宁沈阳110136

出　　处：《润滑与密封》2012年第11期35-38,共4页Lubrication Engineering

基　　金：航空科学基金资助项目(2008ZG54024);中央财政支持地方高校发展专项资金项目(2010年)

摘　　要：航空发动机的磨损机制十分复杂且受诸多不确定因素影响,传统预测方法难以对其磨损趋势进行有效预测。提出一种结构最优化RBF(径向基函数)网络预测模型,采用改进的粒子群算法同时优化模型嵌入维数、核函数宽度及训练误差目标值,实现了RBF网络预测模型最优结构的自动获取。将该方法用于某型航空发动机润滑油金属含量预测,并与传统自回归模型对比,结果证明了该方法的有效性及优越性。The wear mechanism of aero-engine is complex and affected by many complicated factors, therefore traditional method is difficult to forecast the wear trend effectively. An optimal RBF（ radial basis function）neural network forecasting model was put forward. An improved PSO was used to optimize the network embedded dimension, the kernel function width, and the training error, in order to obtain the optimum structure of RBF network forecasting model automatically. The method was applied to forecast the metal content in an aero-engine lubricating oil,and the forecasted result was compared with that of traditional regression model. The superiority and effectiveness of the new method was validated.

关 键 词：航空发动机 金属含量 神经网络 粒子群算法 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] V263.5[自动化与计算机技术—控制科学与工程]