Parallel Solution of Large-Scale Dynamic Optimization Problems

摘要

This paper presents a decomposition strategy applicable to DAE constrained optimization problems. A common solution method for such problems is to apply a direct transcription method and to solve the resulting non-linear program using an interior point algorithm, where the time to solve the linearized KKT system at each iteration is dominating the total solution time. In the proposed method, the structure of the KKT system resulting from a direct collocation scheme for approximating the DAE constraint is exploited in order to distribute the required linear algebra operations on multiple processors. A prototype implementation applied to benchmark models shows promising results.