On Gradient Computation in Single-shooting Nonlinear Model Predictive Control

Ringset, Ruben Køste; Imsland, Lars; Foss, Bjarne A.

doi:10.3182/20100705-3-be-2011.00102

Cited by 1 publication

(2 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The rest of this section treats calculation of the sensitivity matrix by finite differences, and by forward ODE sensitivity integration. We remark that one could also use adjoint sensitivity methods for calculating the desired NMPC gradients (Jørgensen, 2007;Ringset et al, 2010). However, for NMPC problems with a significant number of constraints, this is likely to be less efficient than forward methods.…”

Section: The Sensitivity (Step/impulse Response) Matrixmentioning

confidence: 99%

See 1 more Smart Citation

odel-Based Optimizing Control and Estimation Using Modelica Model

Imsland¹,

Kittilsen²,

Schei³

2010

MIC

View full text Add to dashboard Cite

This paper reports on experiences from case studies in using Modelica/Dymola models interfaced to control and optimization software, as process models in real time process control applications. Possible applications of the integrated models are in state-and parameter estimation and nonlinear model predictive control. It was found that this approach is clearly possible, providing many advantages over modeling in low-level programming languages. However, some effort is required in making the Modelica models accessible to NMPC software. Particular consideration is given to implementation of gradient computation for real-time dynamic optimization, where the dynamic models can be Modelica models. Analytical methods for gradient computation based on sensitivity integration are compared to finite difference-based methods. A case study reveals that analytical methods outperform finite difference-methods as the number of inputs and/or input blocks increases.

show abstract

Section: The Sensitivity (Step/impulse Response) Matrixmentioning

confidence: 99%

“…For efficient implementation, it is essential to exploit input blocking in the sensitivity computations. For further discussion about gradient computations, we refer also to Ringset et al (2010).…”

Section: Nmpc Sensitivity Matrix By Sensitivity Integrationmentioning

confidence: 99%