关键词: 时间序列, 自回归模型, 粒子滤波, 动态更新

Abstract:

As to the problem of autoregressive model built on fixed time series and cannot update with the new value, which result in the poor trend adaptability and low prediction precision, a dynamic model parameter optimization method based on the particle filter is proposed. Firstly, particles are sampled from model parameters to describe the state variables. Further, the weights of particles are calculated using Recursive Bayes estimation method, therefore the posterior filtered estimations can be represented by particles and their weights, and then use them to update the AR model along with the new observe value achieve. The proposed method can improve the predict result accuracy of AR model and give the confidence interval of predict result as well. In the end of the article, effectiveness of the method is verified by lithiumion battery data tested in NASAs Ames Research Center.

Key words: time series, auto regression model, particle filter, dynamic update