MACS members Mihai Anitescu (PI) and Cosmin Petra (CO-PI) are the recipients of a Innovative and Novel Computational Impact on Theory and Experiment (INCITE) Award for 2012-2013 offered by the US Department of Energy (DOE) Office of Science. Through the INCITE award program, DOE supports computationally intensive, large-scale research projects of academia, national laboratories and industry by offering large allocations of computer time and supporting resources at the Argonne and Oak Ridge Leadership Computing Facility (LCF) centers.
INCITE Awards are given to projects selected on a competitive, peer review basis and evaluated for computational readiness. Anitescu’s project, named “Optimization of Complex Energy System Under Uncertainty”, was awarded 24 millions core hours on “Intrepid” BG/P (2012 for 10 million hours and 14 million hours for 2013) and is aimed at developing scalable algorithms and software for the optimal operation of the electrical power grid under uncertainty using high-performance computers. The ultimate goal of the research is to simulate full-resolution stochastic optimization power grid models set up for a vast geographical area (Midwest network) and assess the relationship between various facets of complex energy systems such as stochasticity, renewable energy, economic cost, and reliability.
Increasing energy prices, uncertainties associated with energy imports and environmental concerns caused a shift in the federal government’s energy policy in the last several years. Increasing the nation’s energy efficiency and the energy generation output from domestic, environmental-friendly sources became top priorities in the national energy policy. Undoubtedly, the renewable energy, a term that refers to energy that can be replaced (such as biomass and biofuels), or is readily available (such as wind, solar and geo-thermal) is very appealing since it can be produced domestically and have virtually no impact on the environment. For example, the federal government goal is to have 20% of the electricity produced from wind power by 2030. However, unlike fossil-fuel generation, the available amount of renewable energy that can be produced at any given time is uncertain and highly volatile. As a result, a system relying on renewable energy, such as the U.S. electrical power grid, runs the risk of not meeting consumer demand at peak times.
|Figure 1. Strong scaling efficiency plot when solving the relaxation of a unit commitment problem with a 12-hour horizon.
|Figure 2. Electrical power network of the State of Illinois.
Anitescu and Petra’s INCITE project is focused on using high-performance computing to explore optimization under uncertainty as the paradigm for managing uncertainty in the renewable energy supply. The team uses stochastic programming formulations of the decision process (known as economic dispatch and unit commitment) that schedule supply and match demand. The resulting optimization problems are of extreme-scale (billions of decision variables and billions of equality/inequality constraints) and can be solved only by using supercomputers. Specialized algorithms are needed to efficiently use the supercomputers at full or close to full capacity. The algorithmic developments of the team were implemented in PIPS, a software package for solving structured, large-scale optimization problems and resulted in very good parallel efficiency of PIPS on a wide range of high-performance computers (BG/P, BG/Q, Cray XE6/XK7 and Cray XC30). Figure 1 shows the parallel efficiency of PIPS on up to 131,072 cores of BG/P (80% of full machine). To date, the team has demonstrated that, at least on some configurations such as the network of the State of Illinois showed in Figure 2, even 20% wind penetration, can be accommodated without significant reserve increase if using stochastic optimization, a result that would not be achievable with traditional, deterministic formulations.