Complex Electronic System Performance Degradation Prediction Using Incomplete Measurement Data

A performance degradation prediction method for multi-unit system with insufficient measurement data is proposed by integrating data recovering model, hidden Markov model and support vector regression (SVR) model. The development of the model includes three main parts. Part one, a principal component analysis (PCA) model is build based on normal state. Part two, a hidden Markov model(HMM) is trained based on principal data and log-likelihood ratios that the normal state´s HMM give to the life-cycle historical degradation sequence are calculated to evaluate system degradation. Part three, a SVR model is adopted for modeling degradation process. So, when a new sample with missing data comes, following steps will be taken: recover the principal component based on PCA model, calculate the log-likelihood of degradation sequence based on normal HMM, and then predict future degradation with SVR model. A numerical simulation is taken as an example to show the feasibility and validity of the proposed method.