Exploiting localization for faster power system dynamic simulations

Abstract

Abstract-This paper proposes an algorithm for exploiting the localized response of power system components to accelerate dynamic simulations. During the simulation, components marginally participating to the system dynamics are characterized as latent and their dynamic models are replaced by much simpler equivalents. At the same time, components with significant dynamic activity are characterized as active and their original dynamic models are used. Based on the criterion proposed, components switch status between active and latent to increase performance while retaining accuracy. Two realistic test systems, a medium-scale and a large-scale, are used for the performance evaluation of the proposed method.

Algorithmic and computational advances for fast power system dynamic simulations

“…This is a paper accompanying a Panel presentation at the 2014 IEEE PES General Meeting. The work presented here unifies and extends the previous works [3], [4], [5] and [6]. New simulation results are provided and the benefit of combining the three acceleration techniques is investigated.…”

“…The technique employs simple and fast to compute metrics, originating from real-time digital signal processing, to classify each component into active or latent [6]. In brief, the variability of each injector's active and reactive powers (P i , Q i ) is quantified by computing their standard deviations (P std−i , Q std−i ) over a moving time-window.…”

“…The tolerance value ϵ L controls the trade-off between acceleration and accuracy. In practice, a tolerance ϵ L ≤ 0.5 (MW/MVAr) has been shown to provide significant speedup over several time-steps while maintaining high accuracy [6].…”

Dynamic simulations of combined transmission and distribution systems using decomposition and localization

“…When this value gets smaller than a pre-defined tolerance L , DN i is declared latent. To calculate the standard deviation of S Di efficiently, an approximation originating from real-time digital signal processing is used [5].…”

“…More information on the metrics and the effects of parameters λ 1 and λ 2 on the latency criterion can be found in [5].…”