Power System Probabilistic and Security Analysis Using Commodity High Performance Computing Systems

Given the recent large scale integration of stochastic and varying generation and load in power systems, the conventional deterministic power grid monitoring and analysis tools no longer suffice the comprehensive system analysis and assessment. Additionally, the largely varying grid conditions require the merging of conventional offline security analyses into online operations in order to satisfy the unprecedented security requirements. Meanwhile, the performance capability of modern computing platforms continues to experience rapid growth at an exponential rate. Such advances allow the deployment of substantial computing power (supercomputer-class performance from the mid-2000s) at the substation level at negligible cost. However, these advances in hardware performance result from the increasing complexity of the computer architecture, thus they increase the difficulty of fully utilizing the available computational power for a specific application. These challenges and opportunities in both the power system and the computer engineering fields lead to the main theme of this thesis: harnessing the powerful yet complicated commodity computing systems, developing efficient methods and tools to improve power grid situational awareness given the challenges of stochastic energy integration and security assessment requirements in power systems.;This work targets the fundamental computational kernels for the above power system challenges including the probabilistic load flow (PLF) solver for distribution feeders, the probabilistic extension to supervisory control and data acquisition (SCADA) systems, the PLF solver for transmission grids and the AC contingency calculation for transmission grids. We focus on the performance optimization of the related numerical software solutions on commodity computing systems. This work consists of the following parts: 1) a high performance Monte Carlo simulation (MCS) based three phase distribution probabilistic load flow (DPLF) solver for real-time distribution feeder probabilistic analysis and monitoring. We base this solver on the forward-backward sweep load flow methods. We use aggressive code optimization, multi-level parallelization and task-decomposition for real-time applications to build a solver that runs near the peak performance of modern commodity multi-core CPUs and achieves an order-of-magnitude speedup over the existing baseline software. 2) Building upon the optimized DPLF solver, we develop a proof-of-concept distribution system probabilistic monitoring system (DSPMS) which includes the highly optimized MCS based real-time DPLF solver as well as a real-time communication and visualization system. The DSPMS demonstrates the feasibility and benefits of this real-time probabilistic extension to the SCADA systems. 3) With a similar motivation, we develop a high performance MCS based transmission probabilistic load flow (TPLF) solver. Based on the fast decoupled power flow algorithm, we investigate and develop an efficient linear solver and related elementary functions for parallel load flow computations for the real-time MCS solution of TPLF. 4) Based on similar code optimizations and algorithmic level transformations, we develop an accelerated AC contingency calculation (ACCC) solver for the fast and comprehensive steady state security assessment. Based on Woodbury matrix identity, different contingency cases are transformed into the single instruction multiple data (SIMD) computational model. This together with a thread pool scheduler allows our implementation to fully utilize the computing power of a commodity system, and makes comprehensive ACCC efficient and feasible for real-time application on commodity computing systems.;The goal of this research is to achieve supercomputing class performance on inexpensive commodity computing systems for power system applications. These systems can be deployed at the substation level for real-time applications that address the new challenges in power systems. The broader goal is to demonstrate and discuss the necessary steps and potential impacts of computing performance engineering on specific power system problems, and to bridge the challenges and opportunities in both fields.