"Using Parallel Processing in Power Systems Simulations"

Dr. Rajendra Singh (Post-Doctoral Fellow)
Supervisor: Dr. Aniruddha M. Gole and Dr. Peter Graham

Aim of Project

Modeling and optimization of power systems simulations using parallel processing.

Description of Project

Power systems simulations solve optimization problems by evaluating objective functions, which may require considerable compute power and time to obtain satisfactory solution(s). Running such simulations on a single computer (as is commonly done) can be, in many cases, excessively time consuming relative to acceptable time constraints.

These simulations are iterative in that the same optimization algorithm is invoked multiple times to adjust the values of the parameters used to evaluate the objective function so it converges to a minimum or maximum. Depending on how much the optimization algorithm lends itself to parallelism, there is a lot of potential in harnessing the capabilities of clusters of computers to increase the performance of these simulations by running them in parallel at different levels.

EMTDC is at the core of these power systems simulations. EMTDC, as it is now, is designed to run sequentially on a single computer. Problems that require multiple runs of EMTDC must therefore run it one after another (i.e. sequentially).

As mentioned above, depending on the problem space, we can efficiently parallelize simulation problems (calls to EMTDC) by using MPI and/or OpenMP based parallel processing. At present, we are using MPI to run EMTDC on our compute cluster. Each processor core concurrently runs a copy of EMTDC, each with a different parameter set. Using the 40 available cores we obtain close to 40 times speedup for amenable problems.

We are also exploring GPU based parallel computing for finer grain level problems within each EMTDC run to solve the admittance matrix system faster.

Accepted/Published Papers

None available at this time.