报告题目：A Novel algorithm to general nonlinear filtering problems and its implementation

报告人：罗雪

时间：2013年2月28日（星期四）14：00-15：00

地点：理科楼数学系A304会议室

摘要：In this talk, we shall introduce a novel algorithm to solve the most general settings of nonlinear filtering problems. We first rigorously show in mathematics that our algorithm gives a good approximation of the posterior density function of the ``pathwise-robust" DMZ equation under very mild conditions (which essentially say that the growth of the observation |h| is greater than the growth of the drift |f|). Second, we design our algorithm to maintain the most two important features: in real time manner and without memory. Last, we discuss the implementation of our algorithm by using the Hermite spectral method. Its convergence rate with respect to the truncation mode has been shown. We also investigated the technical difficulties of the implementations, including the scaling factor, the translating factor and the moving-window technique. Two nonlinear filtering problems have been numerically solved by our algorithm and compared with the particle filter to illustrate the accuracy and efficiency of our algorithm.