High Speed Transient Stability, multiprocessing solutions
Kwasnicki, Wieslaw T.
MetadataShow full item record
The emergence of parallel processing architectures and fast network computing have opened new opportunities and challenges to apply these recent technologies to solve power system problems. For the transient stability solution applied on parallel processing hardware, a suitable algorithm has been developed to achieve high speed computation for large systems. This new algorithm combines several techniques useful for parallel processing such as: the W-matrix method for solving network equations, Bus Tearing for splitting large network into smaller subsystems, Current Compensation for handling system admittance changes, node re-ordering scheme to improve matrix sparsity, and a load-balancing partitioning scheme for solving one subsystem on many processors operating in parallel. The multiprocessing algorithm has been incorporated in a stand alone version of a High Speed Transient Stability (HSTS) program. This computer program is aimed at implementations on existing hardware of parallel processing computers such as the Real Time Digital Simulator (RTDS), Distributed Processing Systems (DPS), and other multiprocessing computers (multicomputers). A mechanism that coordinates the scheduling of interdependent operations of a parallel application is provided to run a program concurrently on separate processors. Although, specific implementations require specialized software to achieve fast communication, the basic mechanism for synchronization is built in the HSTS program and is based on the Messa e Passing Interface (MPI) software. Two implementations of the HSTS program have been completed and tested to demonstrate accuracy, efficiency, and computational speed of the proposed multiprocessing solution method. In the RTDS parallel processing implementation, it has been shown that solving the transient stability problem can be performed faster if the system is partitioned for solving on many processors operating in parallel instead of solving by one processor. For large power systems, processing of large admittance matrices takes most of the computation time. In the Distributed Processing System implementation it has been shown that using the system splitting method computation time can be effectively reduced but at the expense of increasing communication time. The high communication latency observed in the Local Area Networks may be eliminated or at least significantly improved by the emerging fast network technologies. Alternative software and hardware tools have been designed by other researchers to synthesize groups of computers into a high-performance environment. One such tool is the High Performance Virtual Machine (HPVM) which, according to the report, can deliver a high-performance message communication over high-speed networks with a bandwidth of 80 megabytes per second and a latency under 11 microseconds using the Myrinet interconnect. It is very probable that with the new technologies of fast network and high performance computing, it will be possible to solve the transient stability problem for large systems in real-time using the HSTS program on a scalable cluster of commodity computers.