融合ADMM相关滤波器与序列重要性重采样的时间序列预测  

Time Series Forecast by ADMM Correlated Filter and Sequence Importance Resampling

作　　者：吴刚 朱勇 封磊 王池社 苏守宝 莫晓晖 WU Gang;ZHU Yong;FENG Lei;WANG Chishe;SU Shoubao;MO Xiaohui(School of Computer Engineering,Jinling Institute of Technology,Manjing 211169,China;Nanjing Innovation Centre of ITS,Nanjing 211169,China)

机构地区：[1]金陵科技学院计算机工程学院,江苏南京211169 [2]南京智能交通创新中心,江苏南京211169

出　　处：《昆明理工大学学报（自然科学版）》2019年第6期46-54,共9页Journal of Kunming University of Science and Technology(Natural Science)

基　　金：国家自然科学基金项目(61801199,61540068);江苏省高等学校自然科学研究项目(19KJA510004);南京市科委资助项目(201704002)

摘　　要：针对序列重要性采样方法中粒子权值计算中存在系统误差的问题,提出一种新的时间序列预测方法,首先由交替方向乘子优化计算相关滤波器,再通过对偶优化变量的计算进而获取目标响应图,然后结合响应图与当前状态量的观测共同引导粒子集的准确定位.对比时间序列预测的序列重要性采样、序列重要性重采样和所提ACSIR方法的实验数据,所提方法有效地降低目标状态估计过程带来的误差.横向对比目前主流的时间序列预测方法,在VOT图像数据集上测试的结果也充分表明:所提方法显著提高了对运动目标的预测成功率.对于城市道路上的交通视频流管理和分析领域,借助所提方法可以有效地对路面车辆进行预测与通行轨迹跟踪,进而为城市交通系统分析与交通需求预测提供关键数据支持.A new time series forecasting method is proposed to correct systematic errors in particle weight calculation in sequential importance sampling method. Firstly, the correlation filter is optimized by the alternating direction multiplier. Then the target response graph is obtained by the calculation of dual optimization variables. Finally, the accurate locations of the particle set are guided by the response graph and the observation of the current state variables. On the basis of comparing the experimental data of time series prediction SIS, SIR and the proposed ACSIR method, the results show that the proposed method can effectively reduce the error caused by the target state estimation process. Compared with the current mainstream time series forecasting methods, the test results on VOT image data sets also show that the proposed method significantly improves the successful forecast rate of moving targets. For the field of traffic video stream management and analysis on urban roads, the proposed method can effectively forecast and track road vehicles, and then provide key data and support for analysis of urban traffic system and forecast of Traffic Demand.

关 键 词：时间序列预测 序列重要性重采样 相关滤波器 交替方向乘子方法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]