Prediction Research About Small Sample Failure Data Based on ARMA Model
Jianguo Huang, Hang Luo, Bing Long, Houjun Wang · 2009
In this paper, conditions and methods of ARMA model's establishment and prediction were detailed analyzed, which were based on correlation characteristics of sample failure data. Because model parameters getting from moment estimation (ME) was very rough to small sample, particle swarm optimization (PSO) algorithm was used in maximum likelihood estimation (MLE) to obtain optimal numerical solutions from probability. Actual verification showed that MLE method based on PSO algorithm could make better digital solutions than ME method. Further more, prediction and its 0.95 confidence interval based on ARMA model to small sample failure data were described, which made prediction have much high credibility, and the results of prediction might give an important reference to objects' failure development trend.