China 
Africa 
Explore chapters and articles related to this topic
Sequential Monte Carlo Methods
Published in Marloes Maathuis, Mathias Drton, Steffen Lauritzen, Martin Wainwright, Handbook of Graphical Models, 2018
and so any ψ $ \boldsymbol{\psi } $ -modulated SMC algorithm can be viewed as a generalization of the auxiliary particle filter. [32] present an approach to approximating the optimal sequence of functions iteratively using SMC, adding to a literature on approximating the exact look-ahead procedure [26,45,45]. These auxiliary particle filter ideas involve twisting in some sense the Markov transition densities and potential functions defining the SMC algorithm. A related idea is to twist instead the dynamics of the particle system defined by the SMC algorithm [76].
A particle filtering approach for temperature based prognostics
Published in Stein Haugen, Anne Barros, Coen van Gulijk, Trond Kongsvik, Jan Erik Vinnem, Safety and Reliability – Safe Societies in a Changing World, 2018
These filters can be divided in different types. The commonly known ones are Auxiliary particle filter, Unscented particle filter, Regularized particle filter, Sequential Importance Sampling (SIS) filter and Sampling Importance Resampling (SIR) filter (Arulampalam et al. 2002). In this work a SIR particle filter is used for estimating the RUL of rubber-metal-bearings due to the named advantages and the SIR related improvement of particle degeneracy. Particle degeneracy is a weakness of the classical SIS filter. The SIR particle filter is a further development of the SIS filter and prevents that degeneracy by resampling. All these Monte Carlo based filters use random samples that are called particles to estimate the state of the monitored product in the form of a distribution. Therefore, the samples’ relevance is symbolized by weights. These weights are calculated based on a defined distribution and a comparison of the predicted and the measured values. In the case of degeneracy after little iteration most of the particle weights tend towards zero while only one particle has a bigger weight. That means that only one particle builds the base for the state estimation and the consecutive estimation of the RUL. Nevertheless all particles are still part of the estimation even if their influence on the result tends towards zero. This degeneracy problem can be solved by resampling. Thereby only relevant samples survive which means samples with a higher weight. Those samples build the base for the next prognostics step while the probably irrelevant particles are no longer considered. In that case a smaller variance of samples is used, but the result is more accurate. The RUL prediction is an iterative method. As long as measured data is available the weights can be updated and resampling can be proceeded (Arulampalam et al. 2002, Jouin et al. 2016).
Particle Filtering
Published in Randal Douc, Eric Moulines, David S. Stoffer, Nonlinear Time Series, 2014
Randal Douc, Eric Moulines, David S. Stoffer
The auxiliary particle filter was introduced in the work by Pitt and Shephard (1999). The consistency and asymptotic normality of the auxiliary particle filter is discussed in Douc et al. (2009b) and Johansen and Doucet (2008), which shows that the auxiliary particle filter can be seen as a particular instance of the Feynman-Kac formula. The concentration properties of interacting particle systems are studied in Del Moral et al. (2010). Non-asymptotic bounds for the auxiliary particle filter are given in Douc et al. (2010).
Real-time model for unit-level heating and cooling energy prediction in multi-family residential housing
Published in Journal of Building Performance Simulation, 2021
Sang Woo Ham, Panagiota Karava, Ilias Bilionis, James Braun
Liu-West filter (LW filter) (Liu and West 2001) takes a slightly different approach. The posterior distributions of parameters are approximated by a mixture of multivariate normal distribution (i.e. kernel mixture smoothing) (West 1993). The posterior distributions of the next time step are updated when new data are available by adding small noise to the current posterior distributions for generating new parameters while the posterior distributions of states are obtained through an auxiliary particle filter (Pitt and Shephard 1999). This is similar to the augmented state-space approach for parameter learning (Simon 2006; Radecki and Hencey 2017). However, adding small noise in each time step results in diffusion in posterior distributions. LW filter overcomes this diffusion problem using location shrinkage of posterior samples to their mean to regulate the dispersion of variance of posterior distributions.
Target Tracking based on Improved Unscented Particle Filter with Markov Chain Monte Carlo
Published in IETE Journal of Research, 2018
To solve this problem, the application of particle filter in target tracking is considered in recent years. An important aspect of particle filter is that it is able to handle any functional nonlinearity and system of noise with any distribution [1,2]. Much research has been done on target tracking based on particle filter. In [14], an auxiliary particle filter is used to track a highly maneuvering target. In [15–17], PF and its variants are used to bearings-only target tracking. In [16–18], Particle filter (PF) with the prior proposal distribution is also applied to the bearings-only tracking. In [13,19], the EKF Gaussian approximation is used as the importance distribution for target tracking based on PF. In [9,20–22], the unscented particle filter (UPF) is developed to target tracking, which uses the UKF [13] to generate the importance proposal distribution.
From Least Squares to Signal Processing and Particle Filtering
Published in Technometrics, 2018
Nozer D. Singpurwalla, Nicholas G. Polson, Refik Soyer
Resampling is another approach by which the effects of degeneracy can be reduced. The idea here is to eliminate particles having a small weight and concentrate on particles with a large weight by picking a particle with a probability proportional to its weight. Such a particle filtering process was proposed by Gordon, Salmond, and Smith (1993) in their classic and ground breaking article; it is known as sampling importance resampling (SIR). Whereas the SIR filter resamples particles at the end of an iteration, say at time (t − 1) before an observation yt at t is taken, the auxiliary particle filter (APF) introduced by Pitt and Shephard (1999), employs the knowledge about yt before resampling at (t − 1). This ensures that particles that are likely to be compatible with yt have a good chance of surviving, and in so doing makes the particle filtering process computationally efficient.