Parallel sensitivity analysis for efficient large-scale dynamic optimization

Hartwich, Arndt; Stockmann, Klaus; Terboven, Christian; Feuerriegel, Stefan; Marquardt, Wolfgang

doi:10.1007/s11081-010-9104-4

Cited by 19 publications

(10 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is worthy to note, that these elements are independent and computing of each element needs the objective or constraint function evaluation [16].…”

Section: Sensitivity Calculationsmentioning

confidence: 99%

Direct Shooting Method for Optimal Control of the Highly Nonlinear Differential-Algebraic Systems

Drąg

Styczeń

2015

Recent Advances in Computational Optimization

View full text Add to dashboard Cite

In the paper the optimal control of highly nonlinear differential-algebraic systems (DAEs) is discussed. The direct shooting method is seen as an efficient tool for control of the complex real-life technological processes, where dynamics and conservation laws are presented. To stabilize the optimization algorithm, the multiple shooting method was proposed. The multiple shooting approach introduces new decision variables and constraints to the problem, but it can preserve the stability of the process, the continuity of the differential state trajectories and enables parallel computation of the mathematical model. The conditions for the frequency of shots, to establish the well-conditioned optimization problem, are considered. The proposed method was tested on the mathematical model of the fed-batch fermentor for penicillin production process, which is a highly nonlinear multistage differentialalgebraic system. The numerical simulations were executed in MATLAB environment using Wroclaw Center for Networking and Supercomputing.

show abstract

“…It is worthy to note, that these elements are independent and computing of each element needs the objective or constraint function evaluation [16].…”

Section: Sensitivity Calculationsmentioning

confidence: 99%

Direct Shooting Method for Optimal Control of the Highly Nonlinear Differential-Algebraic Systems

Drąg

Styczeń

2015

Recent Advances in Computational Optimization

View full text Add to dashboard Cite

show abstract

“…Many large‐scale differential equation solvers provide state sensitivities via forward sensitivity analysis or possibly adjoint (reverse) sensitivity analysis, and the particular numerical methods/techniques used to compute this information are highly important in terms of both solution accuracy and speed when using large‐scale chemical process engineering models . Due to the direct presence of the state variables

x (t_{j + 1})

in the continuity constraints, a natural approach to generate the optimization function derivatives is forward sensitivity analysis.…”

Section: Optimization Solution Frameworkmentioning

confidence: 99%

“…In terms of advancing the state of direct dynamic optimization solution techniques such that larger plant‐wide models can be used, recent work has focused on parallel implementations for: (1) the sensitivity computation within the sequential approach; (2) the embedded model and sensitivity solution over each shooting interval within the multiple‐shooting algorithm; and (3) the structured linear algebra of the Karush–Kuhn–Tucker system solution for the interior‐point (IP) NLP algorithm, used within the simultaneous method . Furthermore, with scenario‐based solution strategies for handling uncertainty, many past implementations have been serial with no exploitation of parallelism for the multiple periods/scenarios used.…”

Section: Introductionmentioning

confidence: 99%

Design under uncertainty using parallel multiperiod dynamic optimization

Washington

Swartz

2014

AIChE Journal

View full text Add to dashboard Cite

A technique for optimizing dynamic systems under uncertainty using a parallel programming implementation is developed in this article. A multiple-shooting discretization scheme is applied, whereby each shooting interval is solved using an error-controlled differential equation solver. In addition, the uncertain parameter space is discretized, resulting in a multiperiod optimization formulation. Each shooting interval and period (scenario) realization is completely independent, thus a major focus of this article is on demonstrating potential computational performance improvement when the embedded dynamic model solution of the multiperiod algorithm is implemented in parallel. We assess our parallel multiperiod and multiple-shooting-based dynamic optimization algorithm on two case studies involving integrated plant and control system design, where the objective is to simultaneously determine the size of the process equipment and the control system tuning parameters that minimize cost, subject to uncertainty in the disturbance inputs.

show abstract

“…Vector data, structured and complex, small storage space needed, spatial location accuracy, is difficult to overlay polygon analysis; raster data with simple data organization and attribute obviously take advantage of comprehensive overlay analysis of spatial information such as Space simulation and space overlay [7][8].…”

Section: ) Rasterizationmentioning

confidence: 99%

Implementation of parallel algorithm for rasterization processing

Huang

Jiang

et al. 2012

2012 20th International Conference on Geoinformatics

View full text Add to dashboard Cite

Parallel algorithms can significantly improve the efficiency of rasterization. This paper implements a parallel algorithm for rasterization that is proposed for the fast processing of enormous amounts of vector data based on message passing interface (MPI) library. Results show that the new parallel algorithm can improve the efficiency of rasterization processing.Geographical Information System (GIS) have not only been using in the geographic field, but also been introduced in many other scientific domains. However, in almost of disciplines to quantify the precise, the high performance capabilities are required that the storage and compute operations for some special applications [1]. They are bottlenecks that traditional GIS software does not meet the needs that computational efficiency is increasing.It is an opportunity for enormous amounts of geo-data conversion limited computational performance that hardware based on parallel computer cluster and multi-core CPU is now becoming more and more common. And parallel computing technology, fo rming parallel computing model that fully perform the hardware function, has been rapidly adopted in the many domain with its maturity and successful application in scientific fields [2] . Moreover, many country have began to build a large-scale collaboration and innovation platform about sharing Scientific Research Equipment, research data and knowledge by high-speed network and computer cluster, fo rming Grid Computing Environment and, for instance, European Enabling Grids for E-sciencE (EGEE), U.S.A Open Science Grid (OSG). Under these circumstances, many researchers have to study the new GIS tools or methods in a high-performance computing Environment. In particular, current research has integrated parallel computing technology in the infrastructure and process in dealing with mass geo-data, such as large image orthophotoscopey and space simulation [3].We proposed a practical approach to exploring various parallelization algorithms on an open source GIS file fo rmats library, which considers ease of reading and writing geo-date from files, rather than fo cusing on GIS and operating system, as has been done in other approach. Because of differences in Supported by the National High Technology Research and Development Program of China (Grant No.2011AA 120301) 978-1-4673-1104-5/12/$31.00 ©20 12 IEEE computing platform, programm ing method, and software framework of parallel computing are different from those of a conventional GIS. The data parallel pattern (DPP), the algorithms parallel pattern (APP) is mainly usual classification of parallel programm ing techniques [4] . Most applications using parallel programm ing approach are data partitioning and function partitioning which together constitute the effective parallel process.Rasterization, a geo-data conversion, is computation intensive method whose computational cost increases greatly when it is applied to large amounts of data and high plan imetric resolution. Rasterization is a spatial data conversion and transformat...

show abstract

Parallel sensitivity analysis for efficient large-scale dynamic optimization

Cited by 19 publications

References 21 publications

Direct Shooting Method for Optimal Control of the Highly Nonlinear Differential-Algebraic Systems

Direct Shooting Method for Optimal Control of the Highly Nonlinear Differential-Algebraic Systems

Design under uncertainty using parallel multiperiod dynamic optimization

Implementation of parallel algorithm for rasterization processing

Contact Info

Product

Resources

About