李文峰　许爱强　尹　晋　朱广辉

(1.海军航空工程学院科研部　烟台　264001)(2.92635部队　青岛　266000)



基于ENN和UKF的电子部件剩余使用寿命预测*

李文峰1,2许爱强1尹晋2朱广辉2

(1.海军航空工程学院科研部烟台264001)(2.92635部队青岛266000)

摘要针对部分可观测信息条件下量测噪声未知时非线性滤波剩余寿命预测的问题,提出了一种基于集成神经网络(ENN)和无迹卡尔曼滤波(UKF)的寿命预测方法。首先,结合设备性能退化量测数据,生成状态-观测数据组,并利用bootstrap技术构建多个数据组,采用集成神经网络训练状态-观测数据组,根据推导公式估计量测噪声标准差的最优取值范围;其次,将量测噪声标准差作为未知参数嵌入在无迹卡尔曼滤波寿命预测框架中,实现非线性系统的剩余寿命预测及概率密度分布;最后,选取电子部件锂电池进行寿命预测仿真验证了该方法的有效性和可行性。

关键词无迹卡尔曼滤波; 集成神经网络; 剩余使用寿命预测; 锂离子电池

Residual Useful Life Prediction for Electronic Component Based on ENN and UKF

LI Wenfeng1,2XU Aiqiang1YIN Jin2ZHU Guanghui2

(1. Department of Scientific Research, Naval Aeronautical and Astronautical University, Yantai264001)

(2. No. 92635 Troop, of PLA, Qingdao266000)

AbstractA new method based on the methods of ensemble neural networks and unscented Kalman filter is proposed for predicting the residual useful life based on the unscented Kalman filter with unknown measurement noise under the condition of partially observable information. Firstly, in combination with the equipment performance degradation data, a state of the group of observed data is generated, and the bootstrap technique is used to construct of a plurality of groups of data, the integrated neural network training state observation data sets are adopted to estimate the measurement noise optimal range according to the formula. Secondly, the residual life prediction and probability density distribution of the nonlinear system are realized by embedding the standard error of measurement noise as the unknown parameters into the framework of unscented Kalman filter lifetime prediction. Finally, the validity and feasibility of the proposed method is verified by the simulation of the life of the lithiumion battery.

Key Wordsunscented Kalman filter, ensemble neural networks, residual useful life prediction, lithium-ion battery

Class NumberTN607

1引言

精确的设备剩余使用寿命(Remaining Useful Life,RUL)预测是维修决策的前提,现实条件下,设备或部件通常表现为非线性系统,且因噪声干扰、检测误差及信息不完备导致的不确定性会影响设备或部件的状态估计或RUL预测。能够处理非线性系统的无迹卡尔曼滤波(Unscented Kalman Filter,UKF)算法广泛应用于状态估计、RUL预测、目标跟踪及组合导航等领域[1~6],文献[7]基于一种根据SOC及电流计算库仑效率的方法,采用用自适应无迹卡尔曼滤波算法预测电池SOC;文献[8]针对水下航行器连续非……