Adaptive Mesh Refinement and Error Estimation Method for Optimal Control Using Direct Collocation

TL;DR

This paper introduces an adaptive mesh refinement and error estimation approach for optimal control problems using Legendre-Gauss-Radau collocation, improving solution accuracy and efficiency through dynamic mesh adjustments.

Contribution

It develops a novel adaptive mesh refinement method combined with a relative error estimate based on explicit simulation, enhancing solution accuracy in optimal control problems.

Findings

Improved mesh size reduction compared to previous methods

Enhanced accuracy of optimal control solutions

Demonstrated effectiveness on benchmark problems

Abstract

An adaptive mesh refinement and error estimation method for numerically solving optimal control problems is developed using Legendre-Gauss-Radau direct collocation. In regions of the solution where the desired accuracy tolerance has not been met, the mesh is refined by either increasing the degree of the approximating polynomial in a mesh interval or dividing a mesh interval into subintervals. In regions of the solution where the desired accuracy tolerance has been met, the mesh size may be reduced by either merging adjacent mesh intervals or decreasing the degree of the approximating polynomial in a mesh interval. Coupled with the mesh refinement method described in this paper is a newly developed relative error estimate that is based on the differences between solutions obtained from the collocation method and those obtained by solving initial-value and terminal-value problems in each mesh interval using an interpolated control obtained from the collocation method. Because the error estimate is based on explicit simulation, the solution obtained via collocation is in close agreement with the solution obtained via explicit simulation using the control on the final mesh, which ensures that the control is an accurate approximation of the true optimal control. The method is demonstrated on three examples from the open literature, and the results obtained show an improvement in final mesh size when compared against previously developed mesh refinement methods.