Localization and latency concepts applied to time simulation of large power systems

Abstract

This paper reports on investigations to speed up timedomain simulations of large electric power systems. First, a decomposed scheme is considered which exploits the bordered block diagonal structure of the original Jacobian involved in the Newton iterations, and keeps the resulting "local" Jacobians constant over multiple iterations. Next, a localization technique is considered, allowing to perform less iterations on the system components with lower activity. Finally, a third scheme consists of visiting only a subset of components, identified as active, and skipping the other ones, identified as latent. The first two techniques solve the whole set of equations with the required accuracy, while the third one involves an adjustable degree of approximation. The methods are illustrated on a small system, while preliminary checks of computational savings from a large test system are reported. Additional results deal with the application of the localization and latency techniques to simplified simulation of the detailed model.

Accelerated and Localized Newton Schemes for Faster Dynamic Simulation of Large Power Systems

“…Similar techniques have been used in the simulation of VLSI circuits [14], where components little affected by a disturbance are called latent. This paper extends the earlier publication [15] in various aspects: (i) general sensitivity formula to deal with any latent component, (ii) improved convergence criteria, (iii) extended results from a larger system and (iv) detailed profiling of execution for accurate speed-up assessment, including comparison with the integrated scheme.…”

“…However, neither the A i nor theC i matrices are updated. Regarding item 1), it is seen from (15,16) …”

“…To tackle these issues, it is convenient to exploit the Jacobian BBD structure and decompose (8) as explained next [15].…”

“…In Schemes A and L, the injector systems (13) are factorized and solved with the DGETR BLAS routines for dense matrices, while the sparse solver ma41 from the Harwell library [20] is used to deal with the network system (15,16). The same sparse solver is used for the whole Jacobian in Scheme I.…”

Exploiting localization for faster power system dynamic simulations

“…While this decomposition method is numerically equivalent to an integrated Newton scheme applied on equations (3) and (4), it provides access to the individual injector models and allows their separate treatment [6]. This feature is exploited by the localization algorithm to switch between full and simplified models during the simulation without expensive operations related to matrix computing and factorizing.…”