Highly Parallel Algorithms and Systems for fast Electromagnetic Transient Simulation in Power System

The significant increase of variable energy resources in the power grid, coupled with the substantial growth of electrified vehicles, leads to a more stressed grid with much higher variabilities at the operational stage. Such variabilities together with today’s lack of accurate simulation capabilities lead to significant uncertainties in predicting the dynamics across the grid, which when combined with lower operating reserves lads to a dangerous combination with respect to grid reliability. To address this requires a capability to accurately predict the future dynamic behavior of the grid in a faster than real-time manner for the range of uncertainties, taking all factors, electromechanical, electrical and electromagnetic, into account. Among all of the transient simulations, electromagnetic transient (EMT) simulation is the most powerful tool which could handle various detailed device models and capture very high speed dynamic behavior in the power system. Thus, simulating the EMTs in a faster than real-time, and high accuracy manner is highly demanded. In this thesis, we propose a computational framework to model, convert, and accelerate the grid-level EMT simulations based on a highly parallel inverse-based low-rank approximation approach which is suitable for full-custom VLSI (very large scale integration) techniques. The first technique we propose is to tailor various models of the key elements in the power system into a general wide-band model and break down the EMT problem into a hardware friendly framework for ASIC architecture. Based on our best knowledge, we adopt the nodal analysis formulation to simulate the EMT problem, which could map multiple device models into the framework. The second technique we propose is to tackle the bottleneck of EMT simulation - network solution which takes 80-97% of the computational time. Traditional approaches to solve network equations are based on sparse LU factorization, which is inherently sequential. We propose a highly parallel inverse-based network solution based on a hierarchical low-rank approximation which permits O(N log N)-time matrix-vector multiplication for each network solution time step. Comprehensive numerical studies are conducted on a 39-bus system and a 179-bus system from the literature, and large cases are created from the two systems. The results demonstrate that the proposed approach is up to 2.8x faster than the state-of-the-art sparse LU factorization based network solution, without compromising simulation accuracy. Since our low-rank approximation is highly parallelizable, the further speedup can be realized by hardware accelerators. The third technique we propose is to reduce the time cost of computing, storing, and updating the inverse of the large sparse conductance matrix in EMT. The inverse-based network solution shows a great benefit in speeding up the EMT simulation with the low-rank approximation approach. However, the computation and storage of the inverse of the conductance matrix will be very expensive especially when a fault occurs. We propose a hierarchical computation and modification method which can not only efficiently compute and storage the inverse of the conductance matrix, but also update the inverse by modifying only local sub-matrices to reflect changes in the network, e.g., loss of a line. Experiments on a series of simplified 179-bus Western Interconnection demonstrate the advantages of the proposed methods. The fourth technique we propose is to develop a full-custom application-specific integrated circuit (ASIC) that can be leveraged for accelerating the EMT simulation in power systems based on the class of parallel algorithms we proposed. Numerical results show that the proposed ASIC architecture can simulate a 2-bus system significantly fast. Large-scale optimal ASIC design will be developed by Naga Shiva Sai Pavan Kumar Devarasetti, a master student of Dr. Weiping Shi