Palestra sobre otimização estocástica multiestágio com Oscar Dowson, criador do SDDP.jl

Oscar Dowson, criador do SDDP.jl e aluno do Prof. Philpott, está no Rio e preparou uma apresentação especialmente para o LAMPS. A palestra, que será ministrada em inglês, acontecerá na próxima segunda-feira (25/06) na sala Multimeios do DEE (grupo 401 do Edifício Cardeal Leme) às 14h. O evento é aberto a todos os interessados. Leiam a seguir, um resumo do tema que será abordado.

New directions for multi-stage stochastic optimization

The focus of many SDDP practitioners has always been on the electricity industry and the hydrothermal scheduling problem. Other application areas such as finance and logistics have been examined but to a much lesser extent. However, one area that is almost un-explored from a multistage stochastic programming viewpoint is agriculture.  This is surprising considering the inherent uncertainty in all steps of the agricultural supply chain. In the past, this may have been because solving large multistage stochastic optimization models using SDDP required the user to code an implementation of the algorithm in addition to the model. This hindered the ability of the researcher to rapidly prototype and explore new models. However, the recent creation of generic SDDP libraries such as SDDP.jl provide researchers with the ability to quickly and easily prototype new stochastic programming models without sacrificing solution speed.

To motivate the discussion, we introduce SDDP.jl and discuss the application of multistage stochastic optimization to the problems faced by dairy farmers in New Zealand. These farmers continually make sequential decisions in the face of short- and long-term weather and price uncertainty. We integrate risk-management via contracting and show how risk-averse farmers can forward-sell their milk to reduce their exposure to price risk.

Although our example focuses on New Zealand dairy farms, we explain how the same tools and techniques easily transfer to many other areas of agriculture. This work is particularly relevant as the world faces challenges associated with global warming, a growing population, and an increasing awareness of the environmental impacts of agriculture. In our view, multistage stochastic optimization is an ideal tool to help participants in the agricultural sector farm smarter in the face of uncertainty.