The 21st Statistical Machine Learning Seminar (2014.10.22)

The 21st Statistical Machine Learning Seminar
Date/time: Oct 22 (Wed) 13:45-15:45
Place: Seminar Room 2 (3F, D304),
Institute of Statistical Mathematics (Tachikawa, Tokyo)

Title : The SCIP Optimization Suite – Concepts, Developments, and Applications
Speaker: Gerald Gamrath (ZIB, Germany)

Abstract : We present the SCIP Optimization Suite, a tool for modeling
and solving optimization problems. It is built around the constraint
integer programming framework SCIP, which is one of the fastest MIP
solvers available in source code. We start with a discussion of the
concepts of SCIP and how they allow to solve a wide range of
optimization problems including pseudo-boolean optimization,
scheduling, and non-convex MINLP. Then, we report on current
developments in the SCIP Optimization Suite and present several
real-world applications in which SCIP is used. Thereby, we elaborate
on how these applications and the challenges they bear push forward
the development of SCIP and our research on mixed-integer programming.
Hereby, we lay the focus on a supply chain management project in which
we regard instances coming from a wide range of applications from all
kinds of industry branches which regularly test the boundaries of our
computational possibilities.

Title : Optimizing Battery Load Schedules
Speaker: Inken Gamrath (ZIB, Germany)

Abstract : As the influence of renewable energy grows, also the
flexible storage of energy gains in importance. For the consumers in
a power grid, it may be advantageous to store and release energy at
suitable times. One aspect in this context is the construction of
storage schedules which provide charging and discharging periods while
considering the storage and power grid properties. In this talk, we
present a model that allows for optimizing a storage schedule for this
purpose. Given predicted demand load curves for a certain future time
period and determined energy prices, the nonlinear model provides a
storage schedule. The model accounts for technical constraints, such
as the charging and discharging losses due to physical properties of
the storage technology, as well as economical and operational ones.
Since environment parameters can change suddenly, our aim is to find
good solutions in a short time in order to allow ad hoc
reoptimization. We present a case study where batteries are used as
storage devices, describe different variants to cope with
nonlinearities and to solve the model, compare them with respect to
quality and efficiency, and show exemplary computations.