Performance study on CUDA GPUs for parallelizing the local ensemble transformed Kalman filter algorithm

Blattner, Timothy; Yang, Shiming

doi:10.1002/cpe.1859

Cited by 7 publications

(5 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…GKF-UI algorithm begins with reading the initial conditions at time k = 0 and partitioning H k and G k into submatrices as shown in Equations (8) and (9). The initial value of the submatrix H k−1 is computed using the Kalman gain (Block 3) given in Equation (10). The error covariance matrices In addition, it is worth mentioning that the major applications of the GKF-UI 12 is for the cases that not all the external excitation data is available or cannot be measured; therefore, traditional Kalman filter (KF) algorithms 3 are not utilizable.…”

Section: Of 17mentioning

confidence: 99%

“…9 4. Popular approaches 5,10 usually parallelize the discrete-time Kalman filter algorithm only by parallelizing matrix operations, and they are unable to achieve optimal performance due to the partial parallelization of the algorithm. However, there are few other techniques in the literature that almost completely parallelize the Kalman filter algorithm, 4,11 but only for special cases.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Machine learning based asynchronous computational framework for generalized Kalman filter

Khani

Abdel‐Qader

2023

Concurrency and Computation

View full text Add to dashboard Cite

Summary Although the Kalman filter algorithms are well suited to be executed on most digital systems, they become slow when applied to large‐scale dynamic systems. Therefore, efficient execution of Kalman filter for the time‐critical and large‐scale applications is of the essence. This work aims to address this necessity by developing a novel framework to improve the performance of a generalized Kalman filter with unknown inputs (GKF‐UI) using multithreaded‐multicore processors and machine learning (ML) classification methods. An asynchronous execution model based on OpenMP message‐passing framework is developed and integrated with a novel supervised ML‐based thread classifier for the GKF‐UI algorithm to enhance execution efficiency. The experimental results show that the proposed approach can achieve up to 35.5× speedup over the serial single‐threaded implementations with no losses in the accuracy or changes to the generality of the filter structure. Moreover, this framework can play a significant role in realizations of computational advantages in large‐scale systems as well as for the time‐critical prediction applications.

show abstract

Section: Of 17mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Machine learning based asynchronous computational framework for generalized Kalman filter

Khani

Abdel‐Qader

2023

Concurrency and Computation

View full text Add to dashboard Cite

show abstract

“…Other data-assimilation methods have also been subject to GPU-accelerations. Blattner and Yang [13] give a performance study of a GPU-implementation of the local ensemble transform Kalman filter, Wei and Huang [14] explore a GPU-based implementation of the EKF, and Quinn and Abarbanel [15] present a general path integral Monte Carlo approach applied to a neuron model. They all report massive speed-ups on the order 100-1000 over CPU implementations.…”

Section: Related Workmentioning

confidence: 99%

Massively parallel implicit equal-weights particle filter for ocean drift trajectory forecasting

Holm

Sætra

Leeuwen

2020

Journal of Computational Physics: X

View full text Add to dashboard Cite

Forecasting ocean drift trajectories are important for many applications, including search and rescue operations, oil spill cleanup and iceberg risk mitigation. In an operational setting, forecasts of drift trajectories are produced based on computationally demanding forecasts of three-dimensional ocean currents. Herein, we investigate a complementary approach for shorter time scales by using a recent state-of-the-art implicit equal-weights particle filter applied to a simplified ocean model. To achieve this, we present a new algorithmic design for a data-assimilation system in which all components -including the model, model errors, and particle filter -take advantage of massively parallel compute architectures, such as graphical processing units. Faster computations can enable in-situ and ad-hoc model runs for emergency management, and larger ensembles for better uncertainty quantification. Using a challenging test case with near-realistic chaotic instabilities, we run data-assimilation experiments based on synthetic observations from drifting and moored buoys, and analyse the trajectory forecasts for the drifters. Our results show that even sparse drifter observations are sufficient to significantly improve short-term drift forecasts up to twelve hours. With equidistant moored buoys observing only 0.1% of the state space, the ensemble gives an accurate description of the true state after data assimilation followed by a high-quality probabilistic forecast. * Corresponding author: havard.heitlo.holm@sintef.no 1 arXiv:1910.01031v1 [stat.CO] 2 Oct 2019 which is an offline trajectory model. It reads the ocean current forecasts produced by the ocean circulation models, and uses these to predict drift trajectories. Although OpenDrift is computationally efficient, the ocean circulation models still require access to supercomputers. This paper explores the option of using a state-of-the-art particle filter method applied to a simplified ocean model for efficient drift trajectory forecasting. The aim is to build a data-assimilation system that can run efficiently on commodity-level desktop computers, and also be extendable to supercomputers. We achieve this by using a simplified ocean model and a data-assimilation method that both are able to take advantage of massively parallel accelerator hardware, such as the graphical processing unit (GPU). This work is not intended as a substitute of current operational systems, but as a complementary approach, in which the predicted currents may even be updated with in-situ observations, e.g., during ongoing search and rescue operations. Furthermore, by enabling research models to run on individual desktop and laptop computers, researchers are able to do more rapid prototyping. At the same time, this work will contribute to more efficient simulations also on supercomputers, since all algorithms may be extended to run on multiple GPUs and compute nodes.The paper is organized as follows: We start by reviewing related work relevant for Lagrangian data assimilation with accelerated pa...

show abstract

“…The speedup reached 1386x and the parallel implementation of extended Kalman filter will serve as good reference on real large-scale applications. Blattner [64] utilized the CUDA programming framework in numerical weather prediction by the data assimilation method, local ensemble transformed Kalman filter algorithm. Results show that an improvement of 72.1 × speedup and provide attractive evidence for applying CUDA GPUs to high demanding scientific computation realms.…”

Section: Introductionmentioning

confidence: 99%

Parallel Computing for Obtaining Regional Scale Rice Growth Conditions Based on WOFOST and Satellite Images

Zhao

Liu

et al. 2020

IEEE Access

View full text Add to dashboard Cite

Performance study on CUDA GPUs for parallelizing the local ensemble transformed Kalman filter algorithm

Cited by 7 publications

References 7 publications

Machine learning based asynchronous computational framework for generalized Kalman filter

Machine learning based asynchronous computational framework for generalized Kalman filter

Massively parallel implicit equal-weights particle filter for ocean drift trajectory forecasting

Parallel Computing for Obtaining Regional Scale Rice Growth Conditions Based on WOFOST and Satellite Images

Contact Info

Product

Resources

About