Decentralized Control and Computation of Large Power Systems

Santa Clara University, Santa Clara CA

Investigators

Abstract

Rapidly increasing demand for power in the newly formed competitive environment of open access transmission can dramatically magnify the undesirable consequences of contingencies, and alter the way we look at system security requirements. This concern presents a number of challenges to the computation and control of large power systems. An important problem in this context is the development of new algorithms that will speed up the computation of control parameters and the simulation of transient phenomena through the use of parallel computing. In order to achieve this objective, it is necessary to develop powerful decomposition schemes which can properly identify dominant subsystems and their interconnections. Epsilon decomposition was developed as a flexible universal algorithm with linear complexity that can identify weakly coupled subsystems. This decomposition was successfully used to identify coherent groups of generators in transient stability studies, as well as to enhance the parallelization of load flow calculations. An important goal of the proposed research will be to develop new variants of epsilon decomposition for a variety of numerical and control applications. We will focus on transient stability analysis by waveform relaxation and parallel solutions of large sparse Riccari equations. In parallelizing transient stability simulation, we propose to apply multilevel epsilon decomposition to the Jacobian of the discretized differential-algebraic equations that describe the machine dynamics. The resulting partitioning will then be used to optimally combine Gauss-Jacobi and Gauss-Seidel waveform relaxation. Particular attention will be devoted to scheduling schemes, which establish the order in which subsystems should be processed, and determine which groups of variables can be updated simultaneously. The parallel solution of large, spare Riccati equations requires a different type of decomposition, one which will simultaneously partition the state and input matrices. In this context, we propose to utilize epsilon decomposition as a preconditioner for the iterative solution of the Riccati equation, in conjunction with Krylov subspace techniques. We will also develop optimal schemes for interprocessor communication, which is a prerequisite for efficient parallel computation.