2020-07-31 –, Green Track
This talk introduces a new flexible and extendable parallel time-domain simulation toolbox developed in Julia for the analysis of power system dynamics in large networks. The simulation algorithm adapts a parallel-in-space decomposition scheme to a sequential algorithm to create parallelizable tasks in the numerical solution of the power system analysis problem. Test simulations using a supercomputing cluster show a huge potential for computational speedup with increasing network complexity.
Dynamic simulations are important in the design and operation of power systems in order to ensure grid stability. Traditional simulation tools in research and industry mainly rely on time-domain simulations based on step-by-step numerical integration. The power system, however, has seen an increase in complexity in light of the current operation of large interconnected networks, growth in electricity demand, and the increasing integration of renewable energies. From the power system analysis perspective, the impact of these changes in operating conditions is an increase in computational complexity in the simulation tools applied for stability and control studies. Parallel and distributed computing techniques are frequently applied to improve the computational speed by taking advantage of multi-core processors and cluster computing. However, the analysis methods for time-domain simulations were developed for sequential operation and optimized for running on single-processors, thereby rendering their application for parallel solutions challenging.
This talk introduces a new Julia-based parallel simulation algorithm to address the need for efficient computation methods in power system stability analysis. The parallel algorithm achieves computational efficiency by adapting an inherently sequential power system numerical solution to a parallel solution using a parallel-in-space decomposition scheme and the Julia computing environment. This talk will describe the parallel-in-space technique which is applied to restructure the power system problem in such a way that it can be applied for formulation of a parallel algorithm. The in-space decomposition is based on the Block Bordered Diagonal Formulation (BBDF) to divide the network coefficient matrix into submatrices that can be solved in parallel. The talk will show how optimal balancing of tasks in the parallel solution process is achieved using a multi-level graph partitioning technique, which is extended to the dynamic simulation problem to obtain balanced subnetworks to be solved in parallel and only linked via an interconnect partition to share information at every time step.
Simulation results will be presented using IEEE standard test networks of varying complexity. The results in the parallel simulation toolbox are compared to those obtained from a sequential implementation in order to validate the solution accuracy and to determine performance improvements in terms of computational speedup.
Michael Kyesswa is a Ph.D student at Karlsruhe Institute of Technology. His main areas of research are computational methods for power system analysis, and parallel and real-time simulations.