By Zulqarnain Haider
Dr. Mario Veiga Pereira is a travelling salesman for using O.R. techniques to solve large scale, complex problems. He runs PSR, an energy planning consultancy firm, out of Rio de Janeiro, Brazil. He leads a team of 60 scientists, many with PhDs in optimization, computer science, and machine learning, to provide invaluable services to countries – 80 by his latest count – throughout the world. Dr. Pereira is not just a consultant but a scholar with hundreds of publications in the areas of energy modeling, energy planning, and stochastic optimization.
In the IFORS Distinguished Lecture, Dr. Pereira combined great energy, wit, and humility, and introduced many recent projects and their unique complexities and challenges. He underscored the importance of rapidly changing landscape of energy planning, shaped by factors such as availability of cheap cloud computing, open source revolution in database and computing resources, availability of copious amounts of accessible data, and the introduction of renewables into the energy mix of many countries.
One recurrent theme of his talk was to bring home the complexities of the real-world problems that necessitate a proactive approach and a constant connection with academia to leverage academic research as fast as possible. Being a stochastic optimization researcher himself, he introduced the very recent advances in dealing with nonconvexities in stochastic dual dynamic programming (SDDP) and the incorporation of these advances into the models at PSR.
He emphasized that modeling of energy systems, with many generation, transmission, and distribution resources is an already complex problem. Introduction of renewables has made the planning problem even more complex with spatial-temporal complementarities in energy production from different resources, multiple planning stages, stochasticity of parameters, nonconvex nonlinearities and many, many integer variables.
The final part of his presentation was about the case study of Genesys planning model in U.S. Pacific Northwest. He introduced the complexity of the Genesys problem, which requires, at its minutest planning period of 1 hour, solving 52 million MIPs, a task achieved by harnessing the combined power of 16000 processing cores on 500 servers for close to three hours. He concluded by sharing the Genesys architecture and the great open source database, statistical and optimization resources, such as Cassandra, SPARK, and Julia JuMP, utilized for solving the large planning problem.