基于粒子滤波优化的滚动式时间序列多步预测

针对复杂的应用环境下，时间序列建模不易准确，多步预测精度不高的问题，提出基于粒子滤波（particle filter, PF）优化的滚动式时间序列（roll time series, RTS）多步预测算法(PF_RTS)。采用Box-Jenkins方法对时间序列滚动自适应建模，所建模型作为粒子的状态转移方程，利用粒子滤波算法实时动态修正预测数据，逼近状态的最优估计。本文算法具有自学习能力，适合实时应用。仿真结果表明，本文算法需要的先验知识少，提高了预测的精度。

Multi-step prediction of rolling time series based on particle filter optimization

For complex application environments, it is difficult to get accurate time series modeling and multi-step prediction results. Multi-step prediction of rolling time series based on the particle filter optimization (PF_RTS) is proposed to solve the problem. According to the modeling thoughts of Box-Jenkins, the time series is adaptively modeled. And the model is regarded as the particles’ state transition equation. By using a particle filtering algorithm, the optimal state is estimated and the predicted data are real-time corrected. With self-learning ability, this algorithm is suitable for real time applications. The simulation results show that this method needs less prior knowledge and has a better predictive accuracy.