 |
Center for Nonlinear Studies |
Mathematical programming has proven to be an efficient tool for design and operation of engineered systems; however, engineering and scientific needs continue to push the boundaries of existing mathematical programming tools, often outstripping the capabilities of a single CPU workstation. Furthermore, computer chip manufacturers are no longer focusing on increasing clock speeds, and the "free" performance improvements that we have historically enjoyed will no longer be available, unless we develop algorithms that are capable of utilizing modern parallel architectures. This presentation discusses advances in parallel algorithms for nonlinear programming problems, and introduces our nonlinear optimization work in power grid applications. Successful operation of transmission networks requires effective unit commitment and economicoptimization to determine generator allocation and set points. Furthermore, grid operators must consider the resiliency of their system to unforeseen failures. In this presentation, I will provide an overview of our work on ACOPF and N-1 contingency constrained ACOPF formulations. Fortunately, these stochastic programming formulations are highly structured. We have developed parallel NLP algorithms that are based on interior-point methods. The dominant computational cost of these algorithms, the solution of the linear KKT system at each iteration, can be effectively parallelized with explicit and implicit Schur-complement techniques. I will also describe these parallel architectures and the advances that enable improved scalability of problems with significant first-stage coupling.
Host: Carleton Coffrin