Graph-Based Modeling and Optimization

Creating, solving, analyzing, and deploying optimization models are core activities in power grid research and industrial practice. These optimization models are computationally intensive and simulations can take days to weeks to complete. Simulations are also often performed using simplified formulations because of the inherent intractability of full-resolution physical models and/or because of the astronomical number of operational situations under which models must be tested. Modeling is also challenging because a large number of concurrent simulation instances and large amounts of input/output data need to be simultaneously managed and analyzed. Model input data often comes in the form of scenarios, which are large data sets that together describe a particular future grid and/or market condition. In some cases, e.g. when dependent on weather forecasts or the computation of high-impact contingencies, expensive external computational procedures are required just for scenario creation. Future infrastructure and supply chain models are expected to become increasingly complex as they incorporate more  distributed modular resources; as well as interactions between multiple networks. State-of-the-art tools are incapable of dealing with this level of complexity.

To tackle these challenges we are collaborating with Argonne National Laboratory to develop PLASMO (Platform for Scalable Modeling and Optimization). PLASMO is an graph-based software platform that facilities the construction, solution, instantiation, and management of optimization models on high-performance computers. PLASMO’s seeks to 1) drastically accelerate the adoption of HPC in the power systems community, 2) enable cost- and energy-efficient use of HPC resources, and 3)  bridge the gap between cutting-edge algorithms and application analysts/researchers. 
 
Multi-Level Graph Abstraction